Skip to content

Self-improving AI Employees for everyone

chore(models): curate the global model lists from the OpenRouter registry

trohitg30cfbb7731cd
diff --git a/docs-internal/native_llm_sdk.md b/docs-internal/native_llm_sdk.md
index 3e6a9bc7..fb77037a 100644
--- a/docs-internal/native_llm_sdk.md
+++ b/docs-internal/native_llm_sdk.md
@@ -99,19 +99,21 @@ Source of truth for this list: `server/config/llm_defaults.json` (the `providers
### Provider / model reference table
-`ModelRegistryService` (`server/services/model_registry.py`) manages the per-model constraints below — fetching from OpenRouter for cloud models and from the user's running local server (`ollama.AsyncClient.ps()` / `lmstudio.AsyncClient.llm.list_loaded()`) for Ollama / LM Studio. Falls back to `llm_defaults.json` only when neither source is available.
+`ModelRegistryService` (`server/services/model_registry.py`) manages the per-model constraints below — fetching from OpenRouter for cloud models and from the user's running local server (`ollama.AsyncClient.ps()` / `lmstudio.AsyncClient.llm.list_loaded()`) for Ollama / LM Studio. Falls back to `llm_defaults.json` only when neither source is available. Kimi and Mistral always use the fallback: the registry files their models under `moonshotai/` and `mistralai/`, which never match those provider names, so their `llm_defaults.json` maps are what requests are sized by.
+
+The curated lists and figures in `llm_defaults.json` and `pricing.json` take the OpenRouter snapshot (`config/model_registry.json`) as their ground truth for every provider it lists, with two exceptions noted in those files: Kimi and Mistral keep max-output budgets below the registry's caps, since those caps (about 80-90% of each window) are what `resolve_max_tokens` would send, and Mistral keeps its `-latest` aliases as keys. The hosts (Groq, Cerebras, Sarvam) are not in the snapshot.
| Provider | Key Models | Context | Max Output | Thinking | Temp Range |
|----------|-----------|---------|-----------|----------|------------|
-| **OpenAI** | GPT-6 Astra (+ `-pro`), GPT-5.6 Sol/Terra/Luna (+ `-pro`; default `gpt-5.6-sol`), GPT-5.5/5.4 | 1.05M | 128K | effort | omitted for reasoning models |
+| **OpenAI** | GPT-6 Sol/Luna/Astra (+ `-pro`; default `gpt-6-sol`), GPT-5.6 Sol/Terra/Luna (+ `-pro`), GPT-5.5/5.4 | 1.05M | 128K | effort | omitted for reasoning models |
| **OpenAI** | GPT-4.1 | ~1.05M | 32K | none | 0-2 |
-| **Anthropic** | Claude Opus 5 (default `claude-opus-5`), Fable 5.1 / 5, Sonnet 5 | 1M | 128K | adaptive | omitted (`temperature` / `top_p` / `top_k` rejected — `sampling_params_removed`) |
+| **Anthropic** | Claude Opus 5.5 (default `claude-opus-5-5`), Opus 5, Fable 5.1 / 5, Sonnet 5 | 1M | 128K | adaptive | omitted (`temperature` / `top_p` / `top_k` rejected — `sampling_params_removed`) |
| **Anthropic** | Claude Opus 4.8/4.7 | 1M | 128K | adaptive | omitted (`sampling_params_removed`) |
| **Anthropic** | Claude Sonnet 4.6 | 1M | 128K | budget | 0-1 |
| **Anthropic** | Claude Haiku 4.5 | 200K | 64K | budget | 0-1 |
| **Google** | Gemini 3.8-flash (default), 3.7/3.6/3.5-flash, 3.5-flash-lite, 3.1-pro-preview/flash-lite, 3-flash-preview, 2.5-pro/flash/flash-lite | 1M | 64K | budget (`thinking_level` on 3.x when set explicitly) | 0-2 |
-| **xAI** | Grok 4.20/4.20-multi-agent, 4.6, 4.5, 4.3, 3 | 131K-1M | 131K | model/provider dependent | 0-2 |
-| **DeepSeek** | deepseek-flash (default; V4.1-Flash), deepseek-v4.1-flash, deepseek-v4-pro (deepseek-v4-flash is a retired alias served by V4.1-Flash; chat/reasoner discontinued) | 1M | 384K | thinking modes | 0-2 |
+| **xAI** | Grok 4.20/4.20-multi-agent, 4.3, 4.7, 4.6, 4.5 | 500K-2M | 450K-1.8M | model/provider dependent | 0-2 |
+| **DeepSeek** | deepseek-v4.1-flash (default), deepseek-v4-pro, deepseek-v4-flash | 1M | 131K (V4.1 Flash); 384K | thinking modes | 0-2 |
| **Kimi** | kimi-k3 (default), kimi-k2.6, kimi-k2.7-code (+ `-highspeed`) | 1M (K3); 256K (K2) | 131K (K3); 32K/96K (K2) | K2 provider default explicitly disabled unless requested | K2 fixed 0.6; K3 0-1 |
| **Mistral** | mistral-large/medium/small-latest (Large 3 / Medium 3.5 / Small 4), codestral-latest | 256K | 32K-131K | none | 0-1.5 |
| **Groq** | GPT-OSS-120b/20b, Llama 3.x tiers, qwen3.8-27b + minimax-m2.7 (preview) | 131K-196K | 16K-131K | effort (GPT-OSS), format (Qwen3) | 0-2 |
@@ -285,7 +287,7 @@ Adding a new OpenAI-compatible provider requires a config entry:
```json
"deepseek": {
- "default_model": "deepseek-flash",
+ "default_model": "deepseek-v4.1-flash",
"detection_patterns": ["deepseek"],
"models_endpoint": "https://api.deepseek.com/models",
"base_url": "https://api.deepseek.com",
@@ -352,11 +354,11 @@ Each provider's `chat()` method reads only the fields it supports. The extracted
| Provider | Models | Parameter | Thinking Type | Notes |
|----------|--------|-----------|---------------|-------|
-| **Claude** (adaptive) | `claude-opus-5`, `claude-fable-5*`, `claude-mythos-5`, `claude-sonnet-5`, `claude-opus-4-8`, `claude-opus-4-7` (`adaptive_thinking_models`, prefix-matched) | `thinkingEnabled` only — the budget is ignored | adaptive | Sends `{"type":"adaptive","display":"summarized"}` when enabled; when disabled the field is omitted (`disabled` is itself rejected on Fable/Mythos). `budget_tokens` is a 400 on these generations. No `temperature` / `top_p` / `top_k` is sent at all (`sampling_params_removed`). `anthropic.py:72-92`. |
+| **Claude** (adaptive) | `claude-opus-5*`, `claude-fable-5*`, `claude-mythos-5`, `claude-sonnet-5`, `claude-opus-4-8`, `claude-opus-4-7` (`adaptive_thinking_models`, prefix-matched) | `thinkingEnabled` only — the budget is ignored | adaptive | Sends `{"type":"adaptive","display":"summarized"}` when enabled; when disabled the field is omitted (`disabled` is itself rejected on Fable/Mythos). `budget_tokens` is a 400 on these generations. No `temperature` / `top_p` / `top_k` is sent at all (`sampling_params_removed`). `anthropic.py:72-92`. |
| **Claude** (budget) | `claude-sonnet-4-6`, `claude-haiku-4-5` and older 4.x/3.5 | `thinkingBudget` (1024-16000 tokens) | budget | `{"type":"enabled","budget_tokens":N}`; provider bumps `max_tokens` to `budget + 1024` if it is not already larger. Temperature auto-set to 1. |
| **Gemini** | gemini-3.x, gemini-2.5-pro/flash | `thinkingBudget` (token count); `thinkingLevel` on 3.x when set explicitly | budget | Uses `thinking_budget` API parameter (`thinking_level` only when the user set it — Vertex rejects an unsolicited level on 2.5-era models) |
| **OpenAI** | `o3`, `o4-mini` (reasoning-only; both on the API-shutdown path per `_models_note`, `o1`/`o3`/`o4` prefixes still detected) | `reasoningEffort` (low/medium/high) | effort | Reasoning-only models. Temperature omitted. |
-| **OpenAI** | GPT-5.6 sol/terra/luna (+ `-pro`), GPT-5.5, GPT-5.4 (`thinking_models: ["gpt-5"]`) | `reasoningEffort` (low/medium/high/xhigh) | effort | Hybrid reasoning: can operate with or without thinking. |
+| **OpenAI** | GPT-6 sol/luna/astra and GPT-5.6 sol/terra/luna (+ `-pro`), GPT-5.5, GPT-5.4 (`thinking_models: ["gpt-5", "gpt-6"]`) | `reasoningEffort` (low/medium/high/xhigh) | effort | Hybrid reasoning: can operate with or without thinking. |
| **Groq** | qwen/qwen3.8-27b (`thinking_models: ["qwen3"]`, prefix match) | `reasoningFormat` ('parsed' or 'hidden') | format | 'parsed' returns reasoning, 'hidden' returns only final answer. GPT-OSS models on Groq use `reasoning_effort` instead (`openai.py:671-675`). |
| **Cerebras** | none — `thinking_models` is empty since zai-glm-4.7 left the public endpoints (2026-09) | `thinkingBudget` | budget | Would be sent as `extra_body.thinking_budget` (`openai.py:111-114`) if a thinking model were curated again. The qwen-3-235b variants are shut down upstream. |
diff --git a/docs-internal/node-logic-flows/ai_chat_models/anthropicChatModel.md b/docs-internal/node-logic-flows/ai_chat_models/anthropicChatModel.md
index ca7dc533..5316f948 100644
--- a/docs-internal/node-logic-flows/ai_chat_models/anthropicChatModel.md
+++ b/docs-internal/node-logic-flows/ai_chat_models/anthropicChatModel.md
@@ -105,7 +105,7 @@ flowchart TD
- **Model ID format**: Anthropic API uses hyphens (`claude-sonnet-4-6`), NOT dots. Dotted IDs (as used by OpenRouter or Gemini) will 404. The frontend model list already uses the correct form, but template-substituted or hand-entered IDs with dots will fail.
- **Thinking budget versus output**: when thinking is enabled and the requested `maxTokens` does not exceed `thinkingBudget`, the native provider raises `max_tokens` to `thinkingBudget + 1024` before calling Anthropic.
- **Temperature forced to 1 when thinking enabled**: regardless of user input.
-- **1M context**: current Claude models (Fable 5, Opus 4.x, Sonnet 4.6) support a 1M context window; configured in `llm_defaults.json` rather than the node.
+- **1M context**: current Claude models (Opus 5.5 / 5, Fable 5.1 / 5, Sonnet 5, Opus 4.8 / 4.7, Sonnet 4.6) support a 1M context window; configured in `llm_defaults.json` rather than the node.
- **Error boundary**: typed Anthropic SDK failures become user-safe `NodeUserError` values in `ChatUnifier` and are re-raised to `BaseNode.execute()`, which produces the standard failure envelope. Unexpected failures are logged and returned by `execute_chat`.
- **No streaming** / **no pricing tracking** on standalone chat nodes.
diff --git a/docs-internal/node-logic-flows/ai_chat_models/deepseekChatModel.md b/docs-internal/node-logic-flows/ai_chat_models/deepseekChatModel.md
index bc670fd4..ce983c67 100644
--- a/docs-internal/node-logic-flows/ai_chat_models/deepseekChatModel.md
+++ b/docs-internal/node-logic-flows/ai_chat_models/deepseekChatModel.md
@@ -11,7 +11,7 @@
## Purpose
-DeepSeek V4.1 models (`deepseek-flash`, `deepseek-v4.1-flash`, `deepseek-v4-pro`); `deepseek-v4-flash` is a retired alias still served by V4.1-Flash, and `deepseek-chat` / `deepseek-reasoner` were discontinued in July 2026. Uses the OpenAI-compatible DeepSeek endpoint via the `services/llm/providers` layer (native path). The `ChatModelBase.chat` operation calls `AIService.execute_chat`. The node's registry description was corrected to the V4.1 line in September 2026; `llm_defaults.json` stays the source of truth for the curated ids.
+DeepSeek V4 models: `deepseek-v4.1-flash` (the default), `deepseek-v4-pro` and the earlier `deepseek-v4-flash`. Uses the OpenAI-compatible DeepSeek endpoint via the `services/llm/providers` layer (native path). The `ChatModelBase.chat` operation calls `AIService.execute_chat`. `llm_defaults.json` holds the curated ids, which follow the OpenRouter snapshot (`model_registry.json`); `deepseek-flash`, another name DeepSeek's API accepts for V4.1-Flash, left the list on 2026-09-26 because the snapshot does not list it.
## Inputs (handles)
@@ -25,7 +25,7 @@ DeepSeek V4.1 models (`deepseek-flash`, `deepseek-v4.1-flash`, `deepseek-v4-pro`
|------|------|---------|----------|---------------------|-------------|
| `prompt` | string | `""` | yes | - | User message |
| `system_prompt` | string | `""` | no | - | System prompt |
-| `model` | string | `""` (injected) | no | - | `deepseek-flash` / `deepseek-v4.1-flash` / `deepseek-v4-pro`; `deepseek-v4-flash` retired alias (served by V4.1-Flash) |
+| `model` | string | `""` (injected) | no | - | `deepseek-v4.1-flash` (default) / `deepseek-v4-pro` / `deepseek-v4-flash` |
| `temperature` | number\|null | `null` | no | - | 0-2 |
| `max_tokens` | number\|null | `null` (8-64K) | no | - | 1-200000 |
| `top_p` | number\|null | `1.0` | no | - | |
@@ -99,7 +99,7 @@ flowchart TD
- **`thinkingEnabled=false` does not suppress a trace**: it only means the UI will not highlight one. If the model returns `reasoning_content`, the response still carries it.
- **`thinkingBudget` has no effect**: DeepSeek reasoning is not budget-configurable; the field is silently ignored.
-- **Context and output**: the curated models use a 1,048,576-token context window and the documented 384,000-token output ceiling. `deepseek-v4-flash` is a retired name DeepSeek still accepts and serves with V4.1-Flash at the Flash price.
+- **Context and output**: all three curated models have a 1,048,576-token context window. The output ceiling is 131,072 tokens for `deepseek-v4.1-flash` and 384,000 for `deepseek-v4-pro` and `deepseek-v4-flash` (OpenRouter snapshot, 2026-09-26).
- **OpenAI-compatible but not OpenAI**: features like `response_format: json_object` have subtly different behavior.
- **Error boundary**: typed OpenAI SDK failures become user-safe `NodeUserError` values in `ChatUnifier` and are re-raised to `BaseNode.execute()`, which produces the standard failure envelope. Unexpected failures are logged and returned by `execute_chat`.
diff --git a/docs-internal/pricing_service.md b/docs-internal/pricing_service.md
index b36a7c40..244ab9c1 100644
--- a/docs-internal/pricing_service.md
+++ b/docs-internal/pricing_service.md
@@ -68,16 +68,16 @@ Located at `server/config/pricing.json`. User-editable; picked up via `PricingSe
```json
{
"version": "2026.09",
- "last_updated": "2026-09-15",
+ "last_updated": "2026-09-26",
"llm": {
"openai": {
- "gpt-5.6-sol": {"input": 4.00, "output": 20.00, "cache_read": 0.40},
- "gpt-5.6-luna": {"input": 0.20, "output": 1.20, "cache_read": 0.02},
+ "gpt-6-sol": {"input": 2.00, "output": 10.00, "cache_read": 0.20},
+ "gpt-6-luna": {"input": 0.10, "output": 0.50, "cache_read": 0.01},
"_default": {"input": 2.50, "output": 15.00}
},
"anthropic": {
- "claude-opus-5": {"input": 5.00, "output": 25.00, "cache_read": 0.50},
+ "claude-opus-5-5": {"input": 4.00, "output": 20.00, "cache_read": 0.40},
"_default": {"input": 3.00, "output": 15.00, "cache_read": 0.30}
}
},
@@ -149,7 +149,7 @@ Located at `server/config/pricing.json`. User-editable; picked up via `PricingSe
| Section | Purpose |
|---------|---------|
-| `llm` | Per-model pricing in USD per million tokens (MTok). Supports `input`, `output`, `cache_read`, `reasoning` |
+| `llm` | Per-model pricing in USD per million tokens (MTok). Supports `input`, `output`, `cache_read`, `reasoning`. `input` and `output` follow the OpenRouter snapshot (`config/model_registry.json`) for every model it lists; it carries no cache price, so `cache_read` keeps each vendor's published cache discount applied to that input price. A lookup tries the exact id, then the first key the id starts with, then the first key it contains, so list a longer id (`gemini-3.5-flash-lite`) before its prefix (`gemini-3.5-flash`) |
| `api` | Per-service pricing in USD per request/resource. Keys starting with `_` are metadata |
| `operation_map` | Maps handler action names to pricing operation keys |
| `url_patterns` | Regex patterns for automatic HTTPX tracking. `count_path` extracts resource count from response JSON |
diff --git a/docs-internal/workflow-schema.md b/docs-internal/workflow-schema.md
index b3d830f5..72f9f20b 100644
--- a/docs-internal/workflow-schema.md
+++ b/docs-internal/workflow-schema.md
@@ -91,8 +91,8 @@ A workflow JSON document contains:
- `openrouterChatModel` - OpenRouter unified API (200+ models)
- `groqChatModel` - Groq ultra-fast inference (Llama, Qwen3, GPT-OSS)
- `cerebrasChatModel` - Cerebras custom AI hardware (Llama, Qwen)
-- `deepseekChatModel` - DeepSeek V4.1 (deepseek-flash / deepseek-v4.1-flash / deepseek-v4-pro; deepseek-v4-flash is a retired alias served by V4.1-Flash, chat/reasoner discontinued)
-- `kimiChatModel` - Moonshot Kimi K2.6 / K2.5 / K2.7-code
+- `deepseekChatModel` - DeepSeek V4 (deepseek-v4.1-flash, the default / deepseek-v4-pro / deepseek-v4-flash)
+- `kimiChatModel` - Moonshot Kimi K3 (the default) / K2.6 / K2.7-code
- `mistralChatModel` - Mistral Large / Small / Codestral
- `ollamaChatModel` - Local Ollama server (OpenAI-compat via `{provider}_proxy` URL)
- `lmstudioChatModel` - Local LM Studio server (OpenAI-compat via `{provider}_proxy` URL)
diff --git a/server/config/llm_defaults.json b/server/config/llm_defaults.json
index 180274cb..f8ea696c 100644
--- a/server/config/llm_defaults.json
+++ b/server/config/llm_defaults.json
@@ -1,7 +1,7 @@
{
- "version": "2026.09.15",
- "last_updated": "2026-09-15",
- "_description": "LLM provider defaults. popular_models drives the ordered global selector and offline fallback independently of token constraints. POLICY (2026-07-23): popular_models lists only models with >= 1M-token context windows — providers whose models are all below 1M carry an empty list and contribute nothing to the global selector (their models stay reachable via the live per-provider fetch_models dropdown). base_url directly configures the nine OpenAI-compatible native providers; dedicated OpenAI, Anthropic, and Gemini providers use SDK defaults, while OpenRouter has its own gateway default. Runtime proxy credentials can override supported endpoints. Model specs are enriched by model_registry.json (OpenRouter cache).",
+ "version": "2026.09.26",
+ "last_updated": "2026-09-26",
+ "_description": "LLM provider defaults. popular_models drives the ordered global selector and offline fallback independently of token constraints. POLICY (2026-07-23): popular_models lists only models with >= 1M-token context windows — providers whose models are all below 1M carry an empty list and contribute nothing to the global selector (their models stay reachable via the live per-provider fetch_models dropdown). SOURCE (2026-09-26): model_registry.json, the OpenRouter snapshot, is the ground truth for the curated ids, context, max output and prices of every provider whose models it lists; an id it does not list is not curated (Mistral's -latest aliases excepted, see its note). It carries nothing for the hosts (groq, cerebras, sarvam), which keep their own documented values. base_url directly configures the nine OpenAI-compatible native providers; dedicated OpenAI, Anthropic, and Gemini providers use SDK defaults, while OpenRouter has its own gateway default. Runtime proxy credentials can override supported endpoints. Model specs are enriched by model_registry.json at runtime; kimi and mistral are the exception, because the registry files their models under moonshotai/ and mistralai/, so their maps here are what requests use.",
"agent": {
"_description": "Agent execution defaults. recursion_limit and compaction.ratio are sourced from the environment (AGENT_RECURSION_LIMIT, COMPACTION_RATIO via core.config.Settings) — these JSON values are last-resort fallbacks when the env vars aren't loaded (e.g. one-off CLI scripts that bypass Settings). default_temperature stays JSON-resident since it has no env override yet.",
@@ -15,8 +15,8 @@
"providers": {
"openai": {
"display_name": "OpenAI",
- "default_model": "gpt-5.6-sol",
- "popular_models": ["gpt-5.6-sol", "gpt-6-astra", "gpt-6-astra-pro", "gpt-5.6-sol-pro", "gpt-5.6-terra", "gpt-5.6-terra-pro", "gpt-5.6-luna", "gpt-5.6-luna-pro", "gpt-5.5", "gpt-5.4", "gpt-4.1"],
+ "default_model": "gpt-6-sol",
+ "popular_models": ["gpt-6-sol", "gpt-6-astra", "gpt-6-astra-pro", "gpt-6-sol-pro", "gpt-6-luna", "gpt-6-luna-pro", "gpt-5.6-sol", "gpt-5.6-sol-pro", "gpt-5.6-terra", "gpt-5.6-terra-pro", "gpt-5.6-luna", "gpt-5.6-luna-pro", "gpt-5.5", "gpt-5.4", "gpt-4.1"],
"detection_patterns": ["gpt", "openai", "o1", "o3", "o4"],
"base_url": "https://api.openai.com/v1",
"models_endpoint": "https://api.openai.com/v1/models",
@@ -26,11 +26,15 @@
"enabled": false,
"_note": "Native image input is emitted only when enabled is true; unknown/absent providers get a text placeholder instead (a rejected image block would poison journaled history). Read by services/llm/media.py provider_supports_vision. False here until the OpenAI request encoder renders image blocks (Responses input_image in function_call_output; Chat Completions hoist-to-user-turn) — flipping it early would hydrate bytes the encoder silently drops. Anthropic ships the first encoder; text-only providers keep the visionAnalyze delegate tool."
},
- "_models_note": "GPT-6 Astra (+ -pro) appear in the 2026-09-11 registry refresh ($10/$50 per MTok, 1.05M context, 128K max output, effort thinking) and are curated as the top tier; the default stays gpt-5.6-sol ($4/$20 per developers.openai.com/api/docs/pricing on 2026-09-15 — terra $2/$12, luna $0.20/$1.20; the router cache under-reports sol at $2/$10), the same price-over-flagship call as anthropic keeping claude-opus-5 over claude-fable-5-1. The rolling -latest aliases (gpt-astra-latest, gpt-sol-latest, gpt-terra-latest, gpt-luna-latest) and the gpt-5.6 alias for Sol are deliberately not curated: real model names only, like gemini. GPT-5.6 family (sol/terra/luna) GA 2026-07-09. gpt-5.5 / gpt-5.4 remain GA as legacy tiers. Dropped: gpt-5.2 (new requests blocked 2026-06-30), o3 (API shutdown 2026-12-11), o4-mini (API shutdown 2026-10-23) — they fall back to _default constraints if still selected before shutdown.",
+ "_models_note": "Ids, context, max output and prices follow the OpenRouter registry (2026-09-26). GPT-6 is the current generation: gpt-6-sol is the default ($2/$10 per MTok), gpt-6-luna the low tier ($0.10/$0.50) and gpt-6-astra the flagship ($10/$50); all have 1.05M context and 128K max output, and each -pro variant is priced like its base model. GPT-6 has no Terra model, so gpt-5.6-terra ($2/$12) stays the middle tier; the rest of the GPT-5.6 family, gpt-5.5, gpt-5.4 and gpt-4.1 stay curated as earlier tiers. The rolling -latest aliases (gpt-astra-latest, gpt-sol-latest, gpt-terra-latest, gpt-luna-latest) are deliberately not curated: real model names only, like gemini. gpt-5.2, o3 and o4-mini are not curated because OpenAI is retiring them; the registry still describes them, so their limits resolve from it.",
"max_output_tokens": {
- "gpt-5.6-sol": 128000,
+ "gpt-6-sol": 128000,
"gpt-6-astra": 128000,
"gpt-6-astra-pro": 128000,
+ "gpt-6-sol-pro": 128000,
+ "gpt-6-luna": 128000,
+ "gpt-6-luna-pro": 128000,
+ "gpt-5.6-sol": 128000,
"gpt-5.6-sol-pro": 128000,
"gpt-5.6-terra": 128000,
"gpt-5.6-terra-pro": 128000,
@@ -42,9 +46,13 @@
"_default": 16384
},
"context_length": {
- "gpt-5.6-sol": 1050000,
+ "gpt-6-sol": 1050000,
"gpt-6-astra": 1050000,
"gpt-6-astra-pro": 1050000,
+ "gpt-6-sol-pro": 1050000,
+ "gpt-6-luna": 1050000,
+ "gpt-6-luna-pro": 1050000,
+ "gpt-5.6-sol": 1050000,
"gpt-5.6-sol-pro": 1050000,
"gpt-5.6-terra": 1050000,
"gpt-5.6-terra-pro": 1050000,
@@ -62,9 +70,9 @@
},
"anthropic": {
"display_name": "Anthropic",
- "default_model": "claude-opus-5",
- "popular_models": ["claude-opus-5", "claude-fable-5-1", "claude-fable-5", "claude-sonnet-5", "claude-opus-4-8", "claude-opus-4-7", "claude-sonnet-4-6"],
- "_models_note": "claude-opus-5 (released 2026-07-24) is the Opus flagship and default: near-Fable intelligence at Opus pricing ($5/$25), 1M context, 128K max output. claude-fable-5-1 (registry 2026-09-03: $10/$50, 1M context, 128K max output) is the most capable generally available model (platform.claude.com/docs/en/models/overview on 2026-09-15: cache reads are 2.5% of input on 5.1 versus 10% elsewhere; claude-sonnet-5 is $2/$10) and supersedes claude-fable-5, which stays curated as the same-priced prior tier. The adaptive-thinking / sampling-removed lists below match by prefix, so the existing claude-fable-5 entries cover 5.1. The claude-opus-5-fast variant (2x pricing) is deliberately not curated here; select it via the live dropdown if needed.",
+ "default_model": "claude-opus-5-5",
+ "popular_models": ["claude-opus-5-5", "claude-fable-5-1", "claude-fable-5", "claude-opus-5", "claude-sonnet-5", "claude-opus-4-8", "claude-opus-4-7", "claude-sonnet-4-6"],
+ "_models_note": "Ids, context, max output and prices follow the OpenRouter registry (2026-09-26), which gives every curated Claude model 1,000,000 tokens of context and 128K max output (Haiku 4.5: 200K and 64K). claude-opus-5-5 is the default: the newest Opus, at $4/$20 per MTok, below claude-opus-5 ($5/$25), which stays curated. claude-fable-5-1 ($10/$50) is the top tier and supersedes claude-fable-5 (same price), also still curated; claude-sonnet-5 is $2/$10. Ids use Anthropic's hyphenated form; the registry's dotted ids (claude-opus-5.5) match them through ModelRegistryService's dot/hyphen variants. The adaptive-thinking / sampling-removed lists below match by prefix, so the claude-opus-5 and claude-fable-5 entries cover 5.5 and 5.1. claude-opus-5-fast is not a registry model and is not curated.",
"detection_patterns": ["claude", "anthropic"],
"base_url": "https://api.anthropic.com/v1",
"models_endpoint": "https://api.anthropic.com/v1/models",
@@ -74,9 +82,10 @@
"enabled": true
},
"max_output_tokens": {
- "claude-opus-5": 128000,
+ "claude-opus-5-5": 128000,
"claude-fable-5-1": 128000,
"claude-fable-5": 128000,
+ "claude-opus-5": 128000,
"claude-opus-4-8": 128000,
"claude-opus-4-7": 128000,
"claude-sonnet-5": 128000,
@@ -85,13 +94,14 @@
"_default": 8192
},
"context_length": {
- "claude-opus-5": 1048576,
- "claude-fable-5-1": 1048576,
- "claude-fable-5": 1048576,
- "claude-opus-4-8": 1048576,
- "claude-opus-4-7": 1048576,
- "claude-sonnet-5": 1048576,
- "claude-sonnet-4-6": 1048576,
+ "claude-opus-5-5": 1000000,
+ "claude-fable-5-1": 1000000,
+ "claude-fable-5": 1000000,
+ "claude-opus-5": 1000000,
+ "claude-opus-4-8": 1000000,
+ "claude-opus-4-7": 1000000,
+ "claude-sonnet-5": 1000000,
+ "claude-sonnet-4-6": 1000000,
"claude-haiku-4-5": 200000,
"_default": 200000
},
@@ -129,7 +139,7 @@
"enabled": false,
"_note": "See openai.vision._note — off until the Gemini encoder renders inline_data parts beside function_response."
},
- "_models_note": "Real (concrete) model names only — no -latest rolling aliases. The curated list (max_output_tokens keys) serves BOTH the Gemini Developer API (AIza keys) and Vertex / Agent Platform (AQ. keys); gemini-* entries live-verified on both endpoints (2026-06-12). Vertex 404s on -latest aliases and its models.list rejects API keys, which is why the dropdown is curated instead of live-fetched. Gemma exception: gemma-4-31b-it + gemma-4-26b-a4b-it are served ONLY on the Developer API (free tier, ai.google.dev/gemma/docs/core/gemma_on_gemini_api, verified 2026-07-11) — Vertex keys must not see them, hence vertex_incompatible_models below. Gemma 4 supports systemInstruction and function calling natively (unlike the gemma-3 era); max output is undocumented, so the curated 8192 is a conservative fallback (registry cache wins when present).",
+ "_models_note": "Real (concrete) model names only — no -latest rolling aliases. The curated list (max_output_tokens keys) serves BOTH the Gemini Developer API (AIza keys) and Vertex / Agent Platform (AQ. keys); gemini-* entries live-verified on both endpoints (2026-06-12). Vertex 404s on -latest aliases and its models.list rejects API keys, which is why the dropdown is curated instead of live-fetched. Gemma exception: gemma-4-31b-it + gemma-4-26b-a4b-it are served ONLY on the Developer API (free tier, ai.google.dev/gemma/docs/core/gemma_on_gemini_api, verified 2026-07-11) — Vertex keys must not see them, hence vertex_incompatible_models below. Gemma 4 supports systemInstruction and function calling natively (unlike the gemma-3 era). Context and max output for every entry follow the OpenRouter registry (2026-09-26), including Gemma's 262K context and 16K / 236K max output.",
"max_output_tokens": {
"gemini-3.8-flash": 65536,
"gemini-3.7-flash": 65536,
@@ -142,8 +152,8 @@
"gemini-2.5-pro": 65536,
"gemini-2.5-flash": 65535,
"gemini-2.5-flash-lite": 65535,
- "gemma-4-31b-it": 8192,
- "gemma-4-26b-a4b-it": 8192,
+ "gemma-4-31b-it": 16384,
+ "gemma-4-26b-a4b-it": 235929,
"_default": 8192
},
"context_length": {
@@ -165,7 +175,7 @@
"_vertex_incompatible_models_note": "Curated models that only exist on the Gemini Developer API, filtered out of fetch_models when the key is a Vertex / Agent Platform key (GeminiProvider._vertex). Gemma is Model Garden self-deploy on Vertex (the 26B MaaS variant needs a different '-maas' id on a different surface), so the plain ids 404 there.",
"vertex_incompatible_models": ["gemma-4-31b-it", "gemma-4-26b-a4b-it"],
"thinking_models": ["gemini-2.5", "gemini-3"],
- "_new_models_note": "gemini-3.8-flash (default) is stable on the Gemini API per ai.google.dev/gemini-api/docs/models/gemini-3.8-flash (2026-09-15: 1,048,576 in / 65,536 out, thinking levels low/medium/high, minimal rejected); gemini-3.7-flash shares the shape. 3.6/3.7/3.8 Flash are $0.75/$3.75 until 2027-01-01, then $1.50/$7.50 (Gemini pricing page); gemini-3.5-flash is $1.50/$9.00, Flash-Lite $0.30/$2.50. The 3.7/3.8 entries are NOT yet live-verified against Vertex / Agent Platform keys like the older entries — drop them into vertex_incompatible_models if a Vertex 404 shows up. gemini-3.8-live and the transcribe / TTS / image models are different endpoints and stay out of the chat list. gemini-3.5-pro is still not GA on the public API (enterprise Vertex preview only) so it stays out of the curated list.",
+ "_new_models_note": "gemini-3.8-flash is the default (registry 2026-09-26: 1,048,576 in / 65,536 out, $0.75/$3.75, the same as 3.7 and 3.6 Flash; gemini-3.5-flash is $1.50/$9.00, Flash-Lite $0.30/$2.50). It accepts thinking levels low/medium/high and rejects minimal. The 3.7/3.8 entries are NOT yet live-verified against Vertex / Agent Platform keys like the older entries — drop them into vertex_incompatible_models if a Vertex 404 shows up. The image, Lyria and -customtools models are different endpoints or variants and stay out of the chat list, as do the -preview ids of models that went GA (gemini-2.5-pro-preview, gemini-3.1-flash-lite-preview).",
"thinking_type": "budget",
"temperature_range": [0.0, 2.0],
"_incompatible_models_note": "Models that the dropdown should hide because they need a different API endpoint than `generate_content`. antigravity-preview-05-2026 requires the Interactions API; selecting it currently surfaces a `google.genai.errors.APIError`. ChatUnifier reads this list in fetch_models and filters before returning. Drop entries here when the unifier gains routing support for the alternate endpoint.",
@@ -174,7 +184,8 @@
"openrouter": {
"display_name": "OpenRouter",
"default_model": "anthropic/claude-sonnet-5",
- "popular_models": ["anthropic/claude-sonnet-5", "anthropic/claude-opus-5", "openai/gpt-5.6-terra", "openai/gpt-5.6-sol", "openai/gpt-6-astra", "google/gemini-3.8-flash", "google/gemini-3.5-flash", "deepseek/deepseek-v4.1-flash", "moonshotai/kimi-k3", "x-ai/grok-4.20"],
+ "popular_models": ["anthropic/claude-sonnet-5", "anthropic/claude-opus-5.5", "openai/gpt-5.6-terra", "openai/gpt-6-sol", "openai/gpt-6-astra", "google/gemini-3.8-flash", "google/gemini-3.5-flash", "deepseek/deepseek-v4.1-flash", "moonshotai/kimi-k3", "x-ai/grok-4.20"],
+ "_popular_note": "Registry ids verbatim (OpenRouter registry, 2026-09-26), one or two current models per vendor; tests/llm/test_global_model_catalogue.py checks each is in model_registry.json.",
"detection_patterns": ["openrouter"],
"open_world_models": true,
"base_url": "https://openrouter.ai/api/v1",
@@ -255,38 +266,39 @@
"display_name": "xAI",
"default_model": "grok-4.20",
"popular_models": ["grok-4.20", "grok-4.20-multi-agent", "grok-4.3"],
- "_popular_note": "grok-4.6 (registry 2026-09-03, 500K context, $2/$6), grok-4.5 (released 2026-07-08) and grok-3 excluded per the 1M-context policy (500K / 500K / 131072); reachable via live fetch_models. Default stays grok-4.20 (1M) so the default is selectable from the global list.",
+ "_popular_note": "grok-4.7, grok-4.6 and grok-4.5 are excluded per the 1M-context policy (500K each); reachable via live fetch_models. Default stays grok-4.20 (2M) so the default is selectable from the global list.",
"detection_patterns": ["grok", "xai"],
"base_url": "https://api.x.ai/v1",
"models_endpoint": "https://api.x.ai/v1/models",
"api_key_param": "api_key",
"max_tokens_param": "max_tokens",
- "_models_note": "grok-4.20 is documented at 1,000,000 context on docs.x.ai/docs/models/grok-4.20 (2026-09-15; the router cache says 2M, the vendor number wins) at $1.25/$2.50, tiered to $2.50/$5.00 once the prompt passes 200K tokens; the bare grok-4.20 id is accepted alongside the dated grok-4.20-0309-reasoning ids and grok-4.20-multi-agent-0309. grok-4.6 and grok-4.5: 500K context, reasoning effort low/medium/high/xhigh (default high), $2/$6 tiered to $4/$12. Output cap undocumented by xAI — curated 131072 matches the rest of the grok family; registry cache wins when present.",
+ "_models_note": "Ids, context, max output and prices follow the OpenRouter registry (2026-09-26): grok-4.20 and grok-4.20-multi-agent have 2M context and 1.8M max output at $1.25/$2.50 per MTok; grok-4.3 1M and 900K at $1.25/$2.50; grok-4.7 500K and 450K at $1.60/$4.80; grok-4.6 and grok-4.5 500K and 450K at $2/$6. grok-3 is not in the registry and is no longer curated. grok-4.6 and grok-4.5 take reasoning effort low/medium/high/xhigh (default high).",
"max_output_tokens": {
- "grok-4.6": 131072,
- "grok-4.5": 131072,
- "grok-4.20": 131072,
- "grok-4.3": 131072,
- "grok-3": 131072,
+ "grok-4.20": 1800000,
+ "grok-4.20-multi-agent": 1800000,
+ "grok-4.3": 900000,
+ "grok-4.7": 450000,
+ "grok-4.6": 450000,
+ "grok-4.5": 450000,
"_default": 8192
},
"context_length": {
- "grok-4.20": 1000000,
- "grok-4.20-multi-agent": 1000000,
+ "grok-4.20": 2000000,
+ "grok-4.20-multi-agent": 2000000,
+ "grok-4.3": 1000000,
+ "grok-4.7": 500000,
"grok-4.6": 500000,
"grok-4.5": 500000,
- "grok-4.3": 1000000,
- "grok-3": 131072,
"_default": 131072
},
"temperature_range": [0.0, 2.0]
},
"deepseek": {
"display_name": "DeepSeek",
- "default_model": "deepseek-flash",
- "popular_models": ["deepseek-flash", "deepseek-v4.1-flash", "deepseek-v4-pro"],
- "_popular_note": "All three curated ids are 1M-context; deepseek-v4-flash stays in the maps only as the retired alias DeepSeek still accepts.",
- "_models_note": "Per api-docs.deepseek.com (updates + pricing, 2026-09-15): DeepSeek-V4.1-Flash shipped 2026-09-10 as deepseek-flash (native multimodal, 1M context, 384K max output, $0.15/$0.60 per MTok off-peak, cache hit $0.003; peak hours 01:00-04:00 and 06:00-10:00 UTC Mon-Fri double it). deepseek-v4-flash and deepseek-v4-flash-vision-exp are retired names that still resolve to V4.1-Flash at the Flash price, so the default moved to deepseek-flash. deepseek-v4-pro (GA 2026-08-13, kept past 2026-09-14) is 1M / 384K at $0.66/$1.98 off-peak. deepseek-chat and deepseek-reasoner were discontinued three months after the 2026-04-24 V4 launch and are no longer curated. Dated snapshots (-0731 / -0813) and the -latest aliases are deliberately not curated.",
+ "default_model": "deepseek-v4.1-flash",
+ "popular_models": ["deepseek-v4.1-flash", "deepseek-v4-pro"],
+ "_popular_note": "Both curated ids are 1M-context; deepseek-v4-flash, the earlier Flash model, stays in the maps but not in the global list.",
+ "_models_note": "Ids, context, max output and prices follow the OpenRouter registry (2026-09-26): deepseek-v4.1-flash (default) has 1M context and 131K max output at $0.14/$0.42 per MTok; deepseek-v4-pro 1M and 384K at $0.4188/$0.8376; deepseek-v4-flash 1M and 384K at $0.047/$0.0941. deepseek-flash, the other name DeepSeek's API accepts for V4.1-Flash, is not a registry model, so it is no longer curated or the default; a node that still names it is sized by the registry's deepseek-flash-latest alias. Dated snapshots (-0731 / -0813), the -latest aliases and deepseek-v4-flash-vision-exp are deliberately not curated.",
"detection_patterns": ["deepseek"],
"base_url": "https://api.deepseek.com",
"models_endpoint": "https://api.deepseek.com/models",
@@ -294,14 +306,12 @@
"max_tokens_param": "max_tokens",
"supported_params": ["temperature", "max_tokens", "top_p", "frequency_penalty", "presence_penalty", "stop", "seed", "tools", "tool_choice", "response_format"],
"max_output_tokens": {
- "deepseek-flash": 384000,
- "deepseek-v4.1-flash": 384000,
+ "deepseek-v4.1-flash": 131072,
"deepseek-v4-pro": 384000,
"deepseek-v4-flash": 384000,
"_default": 8192
},
"context_length": {
- "deepseek-flash": 1048576,
"deepseek-v4.1-flash": 1048576,
"deepseek-v4-pro": 1048576,
"deepseek-v4-flash": 1048576,
@@ -315,7 +325,7 @@
"default_model": "kimi-k3",
"popular_models": ["kimi-k3"],
"_popular_note": "Only kimi-k3 meets the 1M-context policy (k2.x are 262144); k2.x remain reachable via live fetch_models.",
- "_models_note": "kimi-k3 GA (platform.kimi.ai): 2.8T params, 1M context, vision + tool calling + structured outputs, $3/$15 per MTok ($0.30 cache-hit). Default max_completion_tokens 131072 (configurable up to the full window). platform.kimi.ai/docs/pricing/chat (2026-09-15) lists kimi-k2.7-code ($0.95/$4.00, cache hit $0.19), kimi-k2.7-code-highspeed ($1.90/$8.00) and kimi-k2.6 ($0.95/$4.00, cache hit $0.16) as the remaining k2 tiers; kimi-k2.5 is no longer listed and was dropped from the curated maps.",
+ "_models_note": "Ids, context and prices follow the OpenRouter registry (2026-09-26), which files these models under moonshotai/: kimi-k3 has 1M context at $3/$15 per MTok; kimi-k2.6 ($0.95/$4.00) and kimi-k2.7-code ($0.6562/$3.30) have 262K. max_output_tokens deliberately stays below the registry's caps (943,718 for kimi-k3, 235,929 for the k2 models, about 90% of each window): the registry lookup never matches the kimi provider name, so these numbers are what resolve_max_tokens sends as max_tokens, and a cap that close to the window would leave the prompt a tenth of it. 131072 is Moonshot's own default for kimi-k3. kimi-k2.7-code-highspeed is not a registry model, so it has no limits or price of its own here (it matches kimi-k2.7-code by prefix); the thinking and temperature lists still name it because they describe how Moonshot's API treats it.",
"detection_patterns": ["kimi", "moonshot"],
"base_url": "https://api.moonshot.ai/v1",
"models_endpoint": "https://api.moonshot.ai/v1/models",
@@ -328,14 +338,12 @@
"kimi-k3": 131072,
"kimi-k2.6": 32768,
"kimi-k2.7-code": 98304,
- "kimi-k2.7-code-highspeed": 98304,
"_default": 8192
},
"context_length": {
"kimi-k3": 1048576,
"kimi-k2.6": 262144,
"kimi-k2.7-code": 262144,
- "kimi-k2.7-code-highspeed": 262144,
"_default": 262144
},
"temperature_range": [0.0, 1.0]
@@ -345,7 +353,7 @@
"default_model": "mistral-large-latest",
"popular_models": [],
"_popular_note": "Empty per the 1M-context policy — Mistral tops out at 262144 (large/medium/small). Models remain reachable via live fetch_models.",
- "_models_note": "The -latest aliases resolve (docs.mistral.ai models overview, 2026-09-15) to mistral-medium-3504 (Medium 3.5, 256K), mistral-small-2603 (Small 4, 256K, $0.15/$0.60), mistral-large-2512 (Large 3, 256K, $0.50/$1.50) and codestral-2508 ($0.30/$0.90). Mistral does not publish max-output caps; the curated values are conservative and the registry cache wins when present.",
+ "_models_note": "The curated keys are Mistral's -latest aliases, because the node defaults to one and saved nodes hold them; each takes its context and price from the registry model it names (OpenRouter registry, 2026-09-26, under mistralai/): mistral-large-2512 (Large 3, 262K, $0.50/$1.50), mistral-medium-3-5 (Medium 3.5, 262K, $1.50/$7.50), mistral-small-2603 (Small 4, 262K, $0.15/$0.60) and codestral-2508 (256K, $0.30/$0.90). The registry lookup never matches the mistral provider name, so these maps are what requests use, and max_output_tokens keeps the current budgets rather than the registry's caps of about 80% of the window (209,715): resolve_max_tokens sends this number as max_tokens, and a cap that large would leave the prompt a fifth of the window.",
"detection_patterns": ["mistral", "codestral"],
"base_url": "https://api.mistral.ai/v1",
"models_endpoint": "https://api.mistral.ai/v1/models",
diff --git a/server/config/pricing.json b/server/config/pricing.json
index cb4bb1fe..e8a1b627 100644
--- a/server/config/pricing.json
+++ b/server/config/pricing.json
@@ -1,29 +1,34 @@
{
"version": "2026.09",
- "last_updated": "2026-09-15",
+ "last_updated": "2026-09-26",
+ "_llm_note": "LLM input and output prices follow the OpenRouter registry (model_registry.json, 2026-09-26) for every model it lists, USD per MTok. The registry carries no cache price, so cache_read keeps each vendor's published cache discount, applied to the registry input price. The hosts (groq, cerebras, sarvam) are not in the registry and keep their own documented prices.",
"llm": {
"openai": {
- "_note": "developers.openai.com/api/docs/pricing, 2026-09-15 (standard tier). The -pro variants are not on the pricing page; they mirror their base model per the OpenRouter cache.",
+ "_note": "OpenRouter registry, 2026-09-26. cache_read is 10% of input, OpenAI's cached-input rate.",
"gpt-6-astra": {"input": 10.00, "output": 50.00, "cache_read": 1.00},
"gpt-6-astra-pro": {"input": 10.00, "output": 50.00, "cache_read": 1.00},
- "gpt-5.6-sol": {"input": 4.00, "output": 20.00, "cache_read": 0.40},
- "gpt-5.6-sol-pro": {"input": 4.00, "output": 20.00, "cache_read": 0.40},
+ "gpt-6-sol": {"input": 2.00, "output": 10.00, "cache_read": 0.20},
+ "gpt-6-sol-pro": {"input": 2.00, "output": 10.00, "cache_read": 0.20},
+ "gpt-6-luna": {"input": 0.10, "output": 0.50, "cache_read": 0.01},
+ "gpt-6-luna-pro": {"input": 0.10, "output": 0.50, "cache_read": 0.01},
+ "gpt-5.6-sol": {"input": 2.00, "output": 10.00, "cache_read": 0.20},
+ "gpt-5.6-sol-pro": {"input": 2.00, "output": 10.00, "cache_read": 0.20},
"gpt-5.6-terra": {"input": 2.00, "output": 12.00, "cache_read": 0.20},
"gpt-5.6-terra-pro": {"input": 2.00, "output": 12.00, "cache_read": 0.20},
"gpt-5.6-luna": {"input": 0.20, "output": 1.20, "cache_read": 0.02},
"gpt-5.6-luna-pro": {"input": 0.20, "output": 1.20, "cache_read": 0.02},
"gpt-5.5": {"input": 5.00, "output": 30.00, "cache_read": 0.50},
"gpt-5.4": {"input": 2.50, "output": 15.00},
- "gpt-5.3-chat": {"input": 1.75, "output": 14.00},
"gpt-5.2": {"input": 1.75, "output": 14.00},
"o4-mini": {"input": 1.10, "output": 4.40, "reasoning": 4.40},
"gpt-4.1": {"input": 2.00, "output": 8.00},
"_default": {"input": 2.50, "output": 15.00}
},
"anthropic": {
+ "_note": "OpenRouter registry, 2026-09-26. cache_read is Anthropic's cache-read rate, 10% of input (2.5% on claude-fable-5-1).",
+ "claude-opus-5-5": {"input": 4.00, "output": 20.00, "cache_read": 0.40},
"claude-opus-5": {"input": 5.00, "output": 25.00, "cache_read": 0.50},
- "claude-opus-5-fast": {"input": 10.00, "output": 50.00, "cache_read": 1.00},
"claude-fable-5-1": {"input": 10.00, "output": 50.00, "cache_read": 0.25},
"claude-fable-5": {"input": 10.00, "output": 50.00, "cache_read": 1.00},
"claude-opus-4-8": {"input": 5.00, "output": 25.00, "cache_read": 0.50},
@@ -32,25 +37,25 @@
"claude-sonnet-5": {"input": 2.00, "output": 10.00, "cache_read": 0.20},
"claude-sonnet-4-6": {"input": 3.00, "output": 15.00, "cache_read": 0.30},
"claude-haiku-4-5": {"input": 1.00, "output": 5.00, "cache_read": 0.10},
- "claude-sonnet-4-0": {"input": 3.00, "output": 15.00, "cache_read": 0.30},
- "claude-opus-4-0": {"input": 15.00, "output": 75.00, "cache_read": 1.50},
+ "claude-sonnet-4": {"input": 3.00, "output": 15.00, "cache_read": 0.30},
+ "claude-opus-4-1": {"input": 15.00, "output": 75.00, "cache_read": 1.50},
"_default": {"input": 3.00, "output": 15.00, "cache_read": 0.30}
},
"gemini": {
- "_note": "ai.google.dev/gemini-api/docs/pricing paid tier, 2026-09-15, <=200K prompt tier. 3.6/3.7/3.8 Flash rise to $1.50/$7.50 on 2027-01-01.",
+ "_note": "OpenRouter registry, 2026-09-26. For Gemma these are the registry's paid-route prices; Google's own API serves Gemma free, so Gemma usage through it is over-reported.",
"gemini-3.8-flash": {"input": 0.75, "output": 3.75},
"gemini-3.7-flash": {"input": 0.75, "output": 3.75},
"gemini-3.6-flash": {"input": 0.75, "output": 3.75},
- "gemini-3.5-flash": {"input": 1.50, "output": 9.00},
"gemini-3.5-flash-lite": {"input": 0.30, "output": 2.50},
+ "gemini-3.5-flash": {"input": 1.50, "output": 9.00},
"gemini-3.1-flash-lite": {"input": 0.25, "output": 1.50},
"gemini-3.1-pro-preview": {"input": 2.00, "output": 12.00},
"gemini-3-flash-preview": {"input": 0.50, "output": 3.00},
- "gemini-3-pro-preview": {"input": 2.00, "output": 12.00},
"gemini-2.5-pro": {"input": 1.25, "output": 10.00},
+ "gemini-2.5-flash-lite": {"input": 0.10, "output": 0.40},
"gemini-2.5-flash": {"input": 0.30, "output": 2.50},
- "gemma-4-31b-it": {"input": 0.0, "output": 0.0},
- "gemma-4-26b-a4b-it": {"input": 0.0, "output": 0.0},
+ "gemma-4-31b-it": {"input": 0.09, "output": 0.34},
+ "gemma-4-26b-a4b-it": {"input": 0.0675, "output": 0.225},
"_default": {"input": 0.50, "output": 3.00}
},
"groq": {
@@ -64,7 +69,8 @@
"_default": {"input": 0.15, "output": 0.60}
},
"xai": {
- "_note": "docs.x.ai/docs/models, 2026-09-15, base tier (prompts past 200K tokens are billed at double).",
+ "_note": "OpenRouter registry, 2026-09-26.",
+ "grok-4.7": {"input": 1.60, "output": 4.80},
"grok-4.6": {"input": 2.00, "output": 6.00},
"grok-4.5": {"input": 2.00, "output": 6.00},
"grok-4.3": {"input": 1.25, "output": 2.50},
@@ -81,32 +87,27 @@
"_default": {"input": 1.00, "output": 5.00}
},
"deepseek": {
- "_note": "api-docs.deepseek.com/quick_start/pricing, 2026-09-15, off-peak rates (peak hours 01:00-04:00 and 06:00-10:00 UTC Mon-Fri are double). deepseek-v4-flash is a retired alias served by V4.1-Flash at the Flash price; deepseek-chat / deepseek-reasoner are discontinued.",
- "deepseek-flash": {"input": 0.15, "output": 0.60, "cache_read": 0.003},
- "deepseek-v4.1-flash": {"input": 0.15, "output": 0.60, "cache_read": 0.003},
- "deepseek-v4-flash": {"input": 0.15, "output": 0.60, "cache_read": 0.003},
- "deepseek-v4-pro": {"input": 0.66, "output": 1.98, "cache_read": 0.022},
+ "_note": "OpenRouter registry, 2026-09-26. cache_read keeps DeepSeek's published cache-hit ratio (2% of input on Flash, 3.3% on Pro) applied to the registry input price. deepseek-flash is not a registry model and falls to _default.",
+ "deepseek-v4.1-flash": {"input": 0.14, "output": 0.42, "cache_read": 0.0028},
+ "deepseek-v4-flash": {"input": 0.047, "output": 0.0941, "cache_read": 0.00094},
+ "deepseek-v4-pro": {"input": 0.4188, "output": 0.8376, "cache_read": 0.014},
"_default": {"input": 0.15, "output": 0.60}
},
"kimi": {
- "_note": "platform.kimi.ai/docs/pricing/chat, 2026-09-15; cache_read is the cache-hit input price.",
+ "_note": "OpenRouter registry (under moonshotai/), 2026-09-26. cache_read keeps Moonshot's published cache-hit ratio applied to the registry input price. kimi-k2.7-code-highspeed and the moonshot-v1 models are not registry models; the highspeed id matches kimi-k2.7-code by prefix.",
"kimi-k3": {"input": 3.00, "output": 15.00, "cache_read": 0.30},
- "kimi-k2.7-code": {"input": 0.95, "output": 4.00, "cache_read": 0.19},
- "kimi-k2.7-code-highspeed": {"input": 1.90, "output": 8.00, "cache_read": 0.38},
+ "kimi-k2.7-code": {"input": 0.6562, "output": 3.30, "cache_read": 0.1312},
"kimi-k2.6": {"input": 0.95, "output": 4.00, "cache_read": 0.16},
"kimi-k2-thinking": {"input": 0.60, "output": 2.50, "cache_read": 0.15},
- "moonshot-v1-8k": {"input": 0.20, "output": 2.00},
- "moonshot-v1-32k": {"input": 1.00, "output": 3.00},
- "moonshot-v1-128k": {"input": 2.00, "output": 5.00},
"_default": {"input": 0.95, "output": 4.00}
},
"mistral": {
- "_note": "Large 3 (mistral-large-2512), Small 4 (mistral-small-2603) and Codestral from mistral.ai pricing / docs, 2026-09-15; Medium 3.5 (mistral-medium-3504) is not on the public page and mirrors the OpenRouter cache.",
+ "_note": "OpenRouter registry (under mistralai/), 2026-09-26. Each -latest alias takes the price of the model it names: mistral-large-2512, mistral-medium-3-5, mistral-small-2603 and codestral-2508. mistral-nemo also prices Mistral's open-mistral-nemo id, which contains it.",
"mistral-large-latest": {"input": 0.50, "output": 1.50},
"mistral-medium-latest": {"input": 1.50, "output": 7.50},
"mistral-small-latest": {"input": 0.15, "output": 0.60},
"codestral-latest": {"input": 0.30, "output": 0.90},
- "open-mistral-nemo": {"input": 0.15, "output": 0.15},
+ "mistral-nemo": {"input": 0.019, "output": 0.03},
"_default": {"input": 0.15, "output": 0.60}
},
"sarvam": {
diff --git a/server/nodes/model/deepseek_chat_model/__init__.py b/server/nodes/model/deepseek_chat_model/__init__.py
index e3cc04b7..37ad9bf6 100644
--- a/server/nodes/model/deepseek_chat_model/__init__.py
+++ b/server/nodes/model/deepseek_chat_model/__init__.py
@@ -25,7 +25,7 @@ class DeepseekChatModelNode(ChatModelBase):
display_name = "DeepSeek"
subtitle = "Chat Model"
group = ("model",)
- description = "DeepSeek V4.1 models (deepseek-flash, deepseek-v4-pro) with 1M context"
+ description = "DeepSeek V4 models (deepseek-v4.1-flash, deepseek-v4-pro) with 1M context"
credentials = (DeepSeekCredential,)
Params = DeepseekChatModelParams