Skip to content

Codebase intelligence for AI and humans: code health scores, auto-generated docs, git analytics, dead code detection, and architectural decisions via MCP.

feat(providers): support GPT-6 and refresh model lists and prices (#3868)

Raghav Chamadiyadaf7927dbac4
diff --git a/docs/CHANGELOG.md b/docs/CHANGELOG.md
index e018265f8..6e43abadb 100644
--- a/docs/CHANGELOG.md
+++ b/docs/CHANGELOG.md
@@ -14,6 +14,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
### Changed
- **One provider order everywhere.** The CLI, `repowise serve` chat and jobs, and the MCP server now resolve provider, model and base URL as flag or UI choice, then `.repowise/config.yaml`, then the web UI's global default, then `REPOWISE_PROVIDER` / `REPOWISE_MODEL`, so `config.yaml` now wins over those env vars in the CLI. Choosing a model or base URL for a repo in the web UI writes that repo's `config.yaml`, which makes switching models on a local OpenAI-compatible server such as LM Studio a single edit. A base URL in `config.yaml` now beats `OPENAI_BASE_URL` and the other `*_BASE_URL` variables. The default provider and model saved in the web UI settings now also apply to CLI runs in repos whose `config.yaml` names none. A `config.yaml` base URL receives an API key from outside that repo's `.repowise/.env` only when it is a local address or was confirmed in Settings or with `repowise init`.
+- **Current model lists and prices.** Pickers offer GPT-6 Luna and Sol, Claude Sonnet 5.5 and Opus 5.5, Gemini 3.8 Flash and Gemini 4 Argon. DeepSeek now defaults to `deepseek-flash` and LiteLLM to `groq/openai/gpt-oss-120b`, since the old ids were retired. Cost estimates use current rates, and GPT-6 models get the same reasoning, temperature and chat tool handling as GPT-5.6.
## [0.56.0] - 2026-10-12
diff --git a/docs/agent/CLAUDE_CODE_PROVIDER.md b/docs/agent/CLAUDE_CODE_PROVIDER.md
index 640e2bb96..ad43ca7c4 100644
--- a/docs/agent/CLAUDE_CODE_PROVIDER.md
+++ b/docs/agent/CLAUDE_CODE_PROVIDER.md
@@ -87,14 +87,14 @@ booked into the `llm_costs` ledger (at $0.00) so `repowise costs` shows the run.
whose docstring calls haiku "ample for doc pages". To choose another:
```bash
-repowise init --provider claude_cli --model claude_cli/claude-sonnet-4-6
+repowise init --provider claude_cli --model claude_cli/claude-sonnet-5-5
```
The `claude_cli/` prefix is optional on input — a bare slug works too, and both
round-trip through `.repowise/config.yaml`:
```bash
-repowise init --provider claude_cli --model claude-opus-4-6
+repowise init --provider claude_cli --model claude-opus-5-5
```
### Concurrency
diff --git a/docs/reference/CONFIG.md b/docs/reference/CONFIG.md
index d1900af91..ad314315a 100644
--- a/docs/reference/CONFIG.md
+++ b/docs/reference/CONFIG.md
@@ -779,8 +779,8 @@ export ANTHROPIC_API_KEY="sk-ant-..."
| Model | Notes |
|-------|-------|
-| `claude-sonnet-4-6` | Best balance of quality and cost |
-| `claude-opus-4-6` | Highest quality, higher cost |
+| `claude-sonnet-5-5` | Best balance of quality and cost |
+| `claude-opus-5-5` | Highest quality, higher cost |
| `claude-haiku-5-5` | Default, fastest, lowest cost |
```bash
@@ -858,7 +858,7 @@ Gemini is also the default embedding provider when `GEMINI_API_KEY` is set.
```bash
export DEEPSEEK_API_KEY="..."
-repowise init --provider deepseek --model deepseek-chat
+repowise init --provider deepseek --model deepseek-flash
```
### Kimi
diff --git a/packages/cli/src/repowise/cli/ui/provider_selection.py b/packages/cli/src/repowise/cli/ui/provider_selection.py
index c42b359f6..baa1bbc19 100644
--- a/packages/cli/src/repowise/cli/ui/provider_selection.py
+++ b/packages/cli/src/repowise/cli/ui/provider_selection.py
@@ -732,6 +732,7 @@ _FLAGSHIP_MODEL_TOKENS = (
"opus",
"gpt-4o",
"gpt-5",
+ "gpt-6",
"pro",
"sonnet",
"ultra",
diff --git a/packages/core/src/repowise/core/cost_estimator/pricing.py b/packages/core/src/repowise/core/cost_estimator/pricing.py
index a405a11a3..3ba844832 100644
--- a/packages/core/src/repowise/core/cost_estimator/pricing.py
+++ b/packages/core/src/repowise/core/cost_estimator/pricing.py
@@ -30,25 +30,41 @@ _COST_TABLE_EXACT: dict[str, tuple[float, float]] = {
# OpenAI GPT-5.6 family. Same input rate as nano, cheaper output: near
# enough to alias by eye, which is exactly why it gets a real row.
"gpt-5.6-luna": (0.0002, 0.0012), # $0.20 / $1.20 per MTok
- # Gemini family
+ # OpenAI GPT-6 family — https://developers.openai.com/api/docs/pricing
+ "gpt-6-luna": (0.0001, 0.0005), # $0.10 / $0.50 per MTok
+ "gpt-6-sol": (0.002, 0.010), # $2 / $10 per MTok
+ "gpt-6.1-sol": (0.002, 0.010), # $2 / $10 per MTok
+ "gpt-6-astra": (0.01, 0.05), # $10 / $50 per MTok
+ # Gemini family — https://ai.google.dev/gemini-api/docs/pricing
"gemini-3.1-pro-preview": (0.002, 0.012), # $2 / $12 per MTok
"gemini-3-flash-preview": (0.0005, 0.003), # $0.50 / $3 per MTok
"gemini-3.1-flash-lite-preview": (0.00025, 0.0015), # $0.25 / $1.50 per MTok
- "gemini-3.5-flash-lite": (0.00025, 0.0015), # $0.25 / $1.50 per MTok
- # Anthropic Claude 4.x family
+ "gemini-3.5-flash-lite": (0.0003, 0.0025), # $0.30 / $2.50 per MTok
+ # Standard rates; both run an introductory price until year end (Argon
+ # $2 / $10 to Nov 30, 3.8 Flash $0.75 / $3.75 to Dec 31).
+ "gemini-3.8-flash": (0.0015, 0.0075), # $1.50 / $7.50 per MTok
+ "gemini-4-argon": (0.004, 0.02), # $4 / $20 per MTok
+ # Anthropic Claude 4.x and 5.x families
+ "claude-opus-5-5": (0.004, 0.02), # $4 / $20 per MTok
+ "claude-opus-5": (0.005, 0.025), # $5 / $25 per MTok
+ "claude-sonnet-5-5": (0.002, 0.01), # $2 / $10 per MTok
+ "claude-sonnet-5": (0.002, 0.01), # $2 / $10 per MTok
"claude-opus-4-6": (0.005, 0.025), # $5 / $25 per MTok
"claude-sonnet-4-6": (0.003, 0.015), # $3 / $15 per MTok
"claude-haiku-4-5": (0.001, 0.005), # $1 / $5 per MTok
"claude-haiku-5-5": (0.0001, 0.0005), # $0.10 / $0.50 per MTok, prompts <= 100K
# DeepSeek V4 — https://api-docs.deepseek.com/quick_start/pricing
- "deepseek-v4-flash": (0.00027, 0.00110), # $0.27 / $1.10 per MTok
+ # Peak-hour cache-miss rates; off-peak is half. The legacy flash name is
+ # served by V4.1 Flash at the same price.
+ "deepseek-flash": (0.0003, 0.0012), # $0.30 / $1.20 per MTok
+ "deepseek-v4-flash": (0.0003, 0.0012),
"deepseek-v4-pro": (0.00055, 0.00219), # $0.55 / $2.19 per MTok
"deepseek-chat": (0.00027, 0.00110),
"deepseek-reasoner": (0.00055, 0.00219),
- # Kimi / Moonshot — https://platform.moonshot.cn/docs/pricing/chat
+ # Kimi / Moonshot — https://platform.kimi.ai/docs/pricing/chat-k26
"kimi-for-coding": (0.0006, 0.0024), # $0.60 / $2.40 per MTok
"kimi-k2.5": (0.0006, 0.0024),
- "kimi-k2.6": (0.0006, 0.0024),
+ "kimi-k2.6": (0.00095, 0.004), # $0.95 / $4 per MTok
}
# Prefix fallbacks for unknown variants. No `gpt-5.6` catch-all: the 5.6
@@ -56,6 +72,10 @@ _COST_TABLE_EXACT: dict[str, tuple[float, float]] = {
# (the estimator prices a ``None`` from ``lookup_cost`` at zero), so a guess would be worse
# than the miss it hides.
_COST_TABLE_PREFIX: dict[str, tuple[float, float]] = {
+ "gpt-6-luna": (0.0001, 0.0005),
+ "gpt-6-sol": (0.002, 0.010),
+ "gpt-6.1-sol": (0.002, 0.010),
+ "gpt-6-astra": (0.01, 0.05),
"gpt-5.6-luna": (0.0002, 0.0012),
"gpt-5.4-nano": (0.0002, 0.00125),
"gpt-5.4-mini": (0.00075, 0.0045),
@@ -64,17 +84,20 @@ _COST_TABLE_PREFIX: dict[str, tuple[float, float]] = {
"claude-sonnet": (0.003, 0.015),
"claude-haiku": (0.001, 0.005),
"claude": (0.003, 0.015),
- "gemini": (0.00025, 0.0015),
+ "gemini-3.8-flash": (0.0015, 0.0075),
+ "gemini-4-argon": (0.004, 0.02),
+ "gemini": (0.0003, 0.0025),
# DeepSeek V4 family — https://api-docs.deepseek.com/quick_start/pricing
# deepseek-chat was the V3 alias; V4 Flash/Pro are the current names.
"deepseek-v4-pro": (0.00055, 0.00219),
- "deepseek-v4-flash": (0.00027, 0.00110),
- "deepseek-v4": (0.00027, 0.00110),
+ "deepseek-flash": (0.0003, 0.0012),
+ "deepseek-v4-flash": (0.0003, 0.0012),
+ "deepseek-v4": (0.0003, 0.0012),
"deepseek-chat": (0.00027, 0.00110),
"deepseek-reasoner": (0.00055, 0.00219),
"deepseek": (0.00027, 0.00110),
# Kimi / Moonshot — https://platform.moonshot.cn/docs/pricing/chat
- "kimi-k2.6": (0.0006, 0.0024),
+ "kimi-k2.6": (0.00095, 0.004),
"kimi-k2.5": (0.0006, 0.0024),
"kimi-k2": (0.0006, 0.0024),
"kimi-for-coding": (0.0006, 0.0024),
diff --git a/packages/core/src/repowise/core/generation/cost_tracker.py b/packages/core/src/repowise/core/generation/cost_tracker.py
index ef8505be0..d3b6dde45 100644
--- a/packages/core/src/repowise/core/generation/cost_tracker.py
+++ b/packages/core/src/repowise/core/generation/cost_tracker.py
@@ -23,15 +23,18 @@ log = structlog.get_logger(__name__)
# ---------------------------------------------------------------------------
_PRICING: dict[str, dict[str, float]] = {
- # Anthropic — Opus tier $5/$25, Sonnet $3/$15, Haiku $1/$5 per 1M. The 5 and
- # 4-x entries are the models the session detector surfaces from current
- # Claude Code transcripts. Opus used to be listed at $15/$75 here, which is
- # the Opus 3 rate and overstated every Opus session by 3x.
+ # Anthropic — Opus tier $5/$25, Sonnet $3/$15, Haiku $1/$5 per 1M; the 5.x
+ # Sonnets bill $2/$10 and Opus 5.5 $4/$20. The 5 and 4-x entries are the
+ # models the session detector surfaces from current Claude Code transcripts.
+ # Opus used to be listed at $15/$75 here, which is the Opus 3 rate and
+ # overstated every Opus session by 3x.
+ "claude-opus-5-5": {"input": 4.0, "output": 20.0},
"claude-opus-5": {"input": 5.0, "output": 25.0},
"claude-opus-4-8": {"input": 5.0, "output": 25.0},
"claude-opus-4-7": {"input": 5.0, "output": 25.0},
"claude-opus-4-6": {"input": 5.0, "output": 25.0},
- "claude-sonnet-5": {"input": 3.0, "output": 15.0},
+ "claude-sonnet-5-5": {"input": 2.0, "output": 10.0},
+ "claude-sonnet-5": {"input": 2.0, "output": 10.0},
"claude-sonnet-4-6": {"input": 3.0, "output": 15.0},
"claude-haiku-4-5": {"input": 1.0, "output": 5.0},
# Prompts over 100K tokens bill at $0.50/$2.50; doc-page prompts stay under.
@@ -53,6 +56,12 @@ _PRICING: dict[str, dict[str, float]] = {
# gpt-5 branch prices it at the flagship $2.50/$15 tier, 12.5x over on
# output, with no unknown-model warning to notice it by.
"gpt-5.6-luna": {"input": 0.20, "output": 1.20},
+ # GPT-6 has no family fallback below, so an unlisted variant takes the
+ # flat fallback and warns.
+ "gpt-6-luna": {"input": 0.10, "output": 0.50},
+ "gpt-6-sol": {"input": 2.0, "output": 10.0},
+ "gpt-6.1-sol": {"input": 2.0, "output": 10.0},
+ "gpt-6-astra": {"input": 10.0, "output": 50.0},
"gpt-4o": {"input": 2.5, "output": 10.0},
"gpt-4o-mini": {"input": 0.15, "output": 0.6},
# Google Gemini
@@ -65,16 +74,21 @@ _PRICING: dict[str, dict[str, float]] = {
# Gemini preview / experimental models
"gemini-3.1-flash-lite-preview": {"input": 0.075, "output": 0.30},
"gemini-3-flash-preview": {"input": 0.075, "output": 0.30},
- "gemini-3.5-flash-lite": {"input": 0.25, "output": 1.50},
+ "gemini-3.5-flash-lite": {"input": 0.30, "output": 2.50},
+ # Standard rates; introductory prices run until year end.
+ "gemini-3.8-flash": {"input": 1.50, "output": 7.50},
+ "gemini-4-argon": {"input": 4.0, "output": 20.0},
# DeepSeek — https://api-docs.deepseek.com/quick_start/pricing
- "deepseek-v4-flash": {"input": 0.27, "output": 1.10},
+ # Peak-hour rates; the legacy flash name bills as V4.1 Flash.
+ "deepseek-flash": {"input": 0.30, "output": 1.20},
+ "deepseek-v4-flash": {"input": 0.30, "output": 1.20},
"deepseek-v4-pro": {"input": 0.55, "output": 2.19},
"deepseek-chat": {"input": 0.27, "output": 1.10},
"deepseek-reasoner": {"input": 0.55, "output": 2.19},
# Kimi / Moonshot — https://platform.moonshot.cn/docs/pricing/chat
"kimi-for-coding": {"input": 0.60, "output": 2.40},
"kimi-k2.5": {"input": 0.60, "output": 2.40},
- "kimi-k2.6": {"input": 0.60, "output": 2.40},
+ "kimi-k2.6": {"input": 0.95, "output": 4.00},
}
_FALLBACK_PRICING: dict[str, float] = {"input": 3.0, "output": 15.0}
@@ -128,7 +142,9 @@ def is_local_model(model: str) -> bool:
#: the Sonnet fallback rate, which would undercount the agent savings it earned
#: by ~5x. Ordered longest/most-specific prefix first.
_CLAUDE_FAMILY_PRICING: tuple[tuple[str, dict[str, float]], ...] = (
+ ("claude-opus-5-5", {"input": 4.0, "output": 20.0}),
("claude-opus", {"input": 5.0, "output": 25.0}),
+ ("claude-sonnet-5", {"input": 2.0, "output": 10.0}),
("claude-sonnet", {"input": 3.0, "output": 15.0}),
("claude-haiku", {"input": 1.0, "output": 5.0}),
)
diff --git a/packages/core/src/repowise/core/providers/llm/__init__.py b/packages/core/src/repowise/core/providers/llm/__init__.py
index 993a0abaa..7b3f81428 100644
--- a/packages/core/src/repowise/core/providers/llm/__init__.py
+++ b/packages/core/src/repowise/core/providers/llm/__init__.py
@@ -9,11 +9,11 @@ to instantiate a provider by name — this is the preferred entry point.
response = await provider.generate(system_prompt="...", user_prompt="...")
Built-in providers:
- anthropic — claude-opus-4-6, claude-sonnet-4-6, claude-haiku-4-5
- openai — gpt-5.6-luna, gpt-5.4-mini, gpt-5.4
- gemini — gemini-3.1-flash-lite-preview, gemini-3-flash-preview, gemini-3.1-pro-preview
+ anthropic — claude-haiku-5-5, claude-sonnet-5-5, claude-opus-5-5
+ openai — gpt-5.6-luna, gpt-6-luna, gpt-6-sol
+ gemini — gemini-3.5-flash-lite, gemini-4-argon, gemini-3.8-flash
openrouter — 200+ models via OpenRouter (anthropic/claude-sonnet-4.6, etc.)
- deepseek — deepseek-v4-flash, deepseek-v4-pro via api.deepseek.com
+ deepseek — deepseek-flash, deepseek-v4-pro via api.deepseek.com
kimi — kimi-for-coding, kimi-for-coding-highspeed, and K2.x models
edenai — many vendors (mistral/gpt/claude/gemini) via Eden AI's EU gateway
ollama — local inference (llama3.2, codellama, etc.)
diff --git a/packages/core/src/repowise/core/providers/llm/anthropic.py b/packages/core/src/repowise/core/providers/llm/anthropic.py
index 2793ce71d..53ce5c7b5 100644
--- a/packages/core/src/repowise/core/providers/llm/anthropic.py
+++ b/packages/core/src/repowise/core/providers/llm/anthropic.py
@@ -6,8 +6,8 @@ the normal input price on cache hits.
Recommended models (as of 2026):
- claude-haiku-5-5 — fastest and cheapest (default; ample for doc pages)
- - claude-sonnet-4-6 — best quality/cost ratio
- - claude-opus-4-6 — highest quality, most expensive
+ - claude-sonnet-5-5 — best quality/cost ratio
+ - claude-opus-5-5 — highest quality, most expensive
"""
from __future__ import annotations
diff --git a/packages/core/src/repowise/core/providers/llm/base.py b/packages/core/src/repowise/core/providers/llm/base.py
index 18893405f..efab73ea0 100644
--- a/packages/core/src/repowise/core/providers/llm/base.py
+++ b/packages/core/src/repowise/core/providers/llm/base.py
@@ -41,6 +41,7 @@ ModelOptionSource = Literal["api", "local", "fallback"]
_FIXED_TEMPERATURE_PREFIXES = (
# OpenAI reasoning era
"gpt-5",
+ "gpt-6",
"o1",
"o3",
"o4",
diff --git a/packages/core/src/repowise/core/providers/llm/claude_cli.py b/packages/core/src/repowise/core/providers/llm/claude_cli.py
index c8474149c..5f5da7779 100644
--- a/packages/core/src/repowise/core/providers/llm/claude_cli.py
+++ b/packages/core/src/repowise/core/providers/llm/claude_cli.py
@@ -190,8 +190,8 @@ class ClaudeCliProvider(AgentCliProvider):
)
for slug, recommended, notes in (
("claude-haiku-5-5", True, "fastest; ample for doc pages"),
- ("claude-sonnet-4-6", False, "better prose, slower"),
- ("claude-opus-4-6", False, "highest quality; heaviest on subscription limits"),
+ ("claude-sonnet-5-5", False, "better prose, slower"),
+ ("claude-opus-5-5", False, "highest quality; heaviest on subscription limits"),
)
)
diff --git a/packages/core/src/repowise/core/providers/llm/deepseek.py b/packages/core/src/repowise/core/providers/llm/deepseek.py
index db8ba4fa7..f8c3567c3 100644
--- a/packages/core/src/repowise/core/providers/llm/deepseek.py
+++ b/packages/core/src/repowise/core/providers/llm/deepseek.py
@@ -6,7 +6,7 @@ uses the openai Python SDK with a custom base_url, following the same pattern
as OpenRouterProvider.
Models:
- - deepseek-v4-flash — fast, economical (284B total / 13B active params) [default]
+ - deepseek-flash — fast, economical (V4.1 Flash) [default]
- deepseek-v4-pro — highest quality, full reasoning
"""
diff --git a/packages/core/src/repowise/core/providers/llm/edenai.py b/packages/core/src/repowise/core/providers/llm/edenai.py
index bbc5a9de3..d1a314fe1 100644
--- a/packages/core/src/repowise/core/providers/llm/edenai.py
+++ b/packages/core/src/repowise/core/providers/llm/edenai.py
@@ -10,9 +10,9 @@ base_url, following the same pattern as OpenRouterProvider / DeepSeekProvider.
Models use ``vendor/model`` format:
- mistral/mistral-small-latest fast, economical EU model [default]
- - openai/gpt-4o-mini OpenAI small model
- - anthropic/claude-haiku-4-5 Anthropic Claude Haiku
- - google/gemini-2.5-flash Google Gemini Flash
+ - openai/gpt-latest OpenAI's current GPT
+ - anthropic/claude-sonnet-latest Anthropic's current Sonnet
+ - vertex/gemini-flash-latest Google's current Gemini Flash
The live catalogue is public and needs no authentication at
https://api.edenai.run/v3/models.
diff --git a/packages/core/src/repowise/core/providers/llm/gemini.py b/packages/core/src/repowise/core/providers/llm/gemini.py
index a15a1ce47..6b34b1897 100644
--- a/packages/core/src/repowise/core/providers/llm/gemini.py
+++ b/packages/core/src/repowise/core/providers/llm/gemini.py
@@ -5,7 +5,8 @@ Runs the synchronous SDK call in a thread pool to avoid blocking asyncio.
Recommended models:
- gemini-3.5-flash-lite — fast + cheap (default)
- - gemini-3-flash-preview — higher quality
+ - gemini-3.8-flash — higher quality
+ - gemini-4-argon — highest quality
"""
from __future__ import annotations
diff --git a/packages/core/src/repowise/core/providers/llm/litellm.py b/packages/core/src/repowise/core/providers/llm/litellm.py
index 76711bf70..6a5b14959 100644
--- a/packages/core/src/repowise/core/providers/llm/litellm.py
+++ b/packages/core/src/repowise/core/providers/llm/litellm.py
@@ -10,7 +10,7 @@ Use this provider for:
LiteLLM model strings use the format: "<provider>/<model>"
- "together_ai/meta-llama/Llama-3-8b-chat-hf"
- - "groq/llama-3.1-70b-versatile"
+ - "groq/openai/gpt-oss-120b"
- "azure/gpt-4o"
- "bedrock/claude-sonnet-4-6"
@@ -288,7 +288,7 @@ class LiteLLMProvider(BaseProvider):
"""LiteLLM proxy provider — 100+ LLMs through a single interface.
Args:
- model: LiteLLM model string (e.g., "groq/llama-3.1-70b-versatile").
+ model: LiteLLM model string (e.g., "groq/openai/gpt-oss-120b").
api_key: API key for the target provider. Some providers read from
environment variables (e.g., GROQ_API_KEY, TOGETHER_API_KEY).
api_base: Optional custom API base URL (e.g., for self-hosted deployments).
diff --git a/packages/core/src/repowise/core/providers/llm/openai.py b/packages/core/src/repowise/core/providers/llm/openai.py
index 1b378bd91..f58e3ea95 100644
--- a/packages/core/src/repowise/core/providers/llm/openai.py
+++ b/packages/core/src/repowise/core/providers/llm/openai.py
@@ -5,9 +5,9 @@ Also works as a base for any OpenAI-compatible API endpoint via the
`base_url` parameter.
Recommended models (as of 2026):
- - gpt-5.6-luna — fastest, cheapest ($0.20/$1.20 per MTok) [default]
- - gpt-5.4-mini — balanced speed and quality ($0.75/$4.50 per MTok)
- - gpt-5.4 — highest quality ($2.50/$15 per MTok)
+ - gpt-5.6-luna — fast, cheap ($0.20/$1.20 per MTok) [default]
+ - gpt-6-luna — cheapest ($0.10/$0.50 per MTok)
+ - gpt-6-sol — highest quality ($2/$10 per MTok)
"""
from __future__ import annotations
@@ -74,7 +74,7 @@ def _model_leaf(model: str) -> str:
def _supports_openai_reasoning_effort(model: str) -> bool:
leaf = _model_leaf(model)
- return leaf.startswith(("gpt-5", "o1", "o3", "o4"))
+ return leaf.startswith(("gpt-5", "gpt-6", "o1", "o3", "o4"))
def _supports_chat_template_thinking_toggle(model: str) -> bool:
@@ -95,7 +95,10 @@ def _openai_supported_reasoning_modes(model: str) -> tuple[ReasoningMode, ...]:
return ("none", "low", "medium", "high")
if leaf.startswith("gpt-5-pro"):
return ("high",)
- if leaf.startswith("gpt-5.6"):
+ if leaf.startswith("gpt-6.1"):
+ # 6.1 dropped `none` as well (verified live on gpt-6.1-sol, 2026-10-15).
+ return ("low", "medium", "high", "xhigh")
+ if leaf.startswith(("gpt-5.6", "gpt-6")):
# 5.6 dropped `minimal` and added `xhigh`. Must stay above the generic
# `gpt-5` branch, which is a prefix of this one and would otherwise win
# and offer `minimal`, which 5.6 rejects, and only on a live call.
@@ -104,7 +107,8 @@ def _openai_supported_reasoning_modes(model: str) -> tuple[ReasoningMode, ...]:
# API rejects it with `unsupported_value`, naming exactly the five
# below. Verified live on both gpt-5.6-luna and gpt-5.6-sol
# (2026-08-15). Family-level rather than per-model on the same
- # evidence: the two variants answered identically.
+ # evidence: the two variants answered identically. gpt-6-luna and
+ # gpt-6-sol answered the same way (2026-10-15).
return ("none", "low", "medium", "high", "xhigh")
if leaf.startswith("gpt-5"):
return ("minimal", "low", "medium", "high")
@@ -175,14 +179,14 @@ LOCAL_SERVER_API_KEY = "not-needed"
def _openai_chat_tool_kwargs(model: str, *, has_tools: bool) -> dict[str, Any]:
"""Return Chat Completions overrides required by tool-enabled models.
- GPT-5.6 models default to a non-none reasoning effort. OpenAI recommends
+ GPT-5.6 and GPT-6 models default to a non-none reasoning effort. OpenAI recommends
the Responses API for reasoning with tool-calling, while this provider's
repository chat loop still uses Chat Completions. Keep the family's normal
reasoning default for generation and only disable it when function tools
are attached. Migrating the shared chat protocol is a separate change.
"""
leaf = _model_leaf(model)
- if has_tools and (leaf == "gpt-5.6" or leaf.startswith("gpt-5.6-")):
+ if has_tools and (leaf in ("gpt-5.6", "gpt-6") or leaf.startswith(("gpt-5.6-", "gpt-6-"))):
return {"reasoning_effort": "none"}
return {}
diff --git a/packages/core/src/repowise/core/providers/llm/openrouter.py b/packages/core/src/repowise/core/providers/llm/openrouter.py
index 854116d4e..5b68c0333 100644
--- a/packages/core/src/repowise/core/providers/llm/openrouter.py
+++ b/packages/core/src/repowise/core/providers/llm/openrouter.py
@@ -7,8 +7,8 @@ No additional pip install required — uses the ``openai`` package.
Popular models:
- google/gemini-3.5-flash-lite — fast + cheap (default)
- - openai/gpt-5.6-luna — OpenAI budget tier
- - anthropic/claude-haiku-4-5 — Anthropic budget tier
+ - openai/gpt-6-luna — OpenAI budget tier
+ - google/gemini-3.8-flash — higher quality
"""
from __future__ import annotations
diff --git a/packages/core/src/repowise/core/providers/llm/specs.py b/packages/core/src/repowise/core/providers/llm/specs.py
index 4ff8a74ac..2cc4ff58d 100644
--- a/packages/core/src/repowise/core/providers/llm/specs.py
+++ b/packages/core/src/repowise/core/providers/llm/specs.py
@@ -114,7 +114,7 @@ _SPECS: tuple[ProviderSpec, ...] = (
impl=f"{_LLM}.gemini:GeminiProvider",
label="Google Gemini",
default_model="gemini-3.5-flash-lite",
- models=("gemini-3.5-flash-lite", "gemini-3.1-flash-lite", "gemini-3.1-pro-preview"),
+ models=("gemini-3.5-flash-lite", "gemini-4-argon", "gemini-3.8-flash"),
api_key_envs=("GEMINI_API_KEY", "GOOGLE_API_KEY"),
base_url_envs=("GEMINI_BASE_URL",),
rate_limit=RateLimitConfig(requests_per_minute=1_000, tokens_per_minute=4_000_000),
@@ -129,7 +129,7 @@ _SPECS: tuple[ProviderSpec, ...] = (
impl=f"{_LLM}.anthropic:AnthropicProvider",
label="Anthropic",
default_model="claude-haiku-5-5",
- models=("claude-haiku-5-5", "claude-sonnet-4-6", "claude-opus-4-6"),
+ models=("claude-haiku-5-5", "claude-sonnet-5-5", "claude-opus-5-5"),
api_key_envs=("ANTHROPIC_API_KEY",),
base_url_envs=("ANTHROPIC_BASE_URL",),
rate_limit=RateLimitConfig(requests_per_minute=2_000, tokens_per_minute=1_600_000),
@@ -143,7 +143,7 @@ _SPECS: tuple[ProviderSpec, ...] = (
impl=f"{_LLM}.openai:OpenAIProvider",
label="OpenAI",
default_model="gpt-5.6-luna",
- models=("gpt-5.6-luna", "gpt-5.4-nano", "gpt-5.4-mini", "gpt-5.4"),
+ models=("gpt-5.6-luna", "gpt-6-luna", "gpt-6-sol"),
api_key_envs=("OPENAI_API_KEY",),
base_url_envs=("OPENAI_BASE_URL",),
keyless_on_loopback=True,
@@ -160,9 +160,9 @@ _SPECS: tuple[ProviderSpec, ...] = (
default_model="google/gemini-3.5-flash-lite",
models=(
"google/gemini-3.5-flash-lite",
+ "openai/gpt-6-luna",
"openai/gpt-5.6-luna",
- "anthropic/claude-haiku-4-5",
- "anthropic/claude-sonnet-5",
+ "google/gemini-3.8-flash",
),
api_key_envs=("OPENROUTER_API_KEY",),
rate_limit=RateLimitConfig(requests_per_minute=1_000, tokens_per_minute=2_000_000),
@@ -175,8 +175,8 @@ _SPECS: tuple[ProviderSpec, ...] = (
picker_rank=3,
impl=f"{_LLM}.deepseek:DeepSeekProvider",
label="DeepSeek",
- default_model="deepseek-v4-flash",
- models=("deepseek-v4-flash", "deepseek-v4-pro"),
+ default_model="deepseek-flash",
+ models=("deepseek-flash", "deepseek-v4-pro"),
api_key_envs=("DEEPSEEK_API_KEY",),
base_url_envs=("DEEPSEEK_BASE_URL",),
rate_limit=RateLimitConfig(requests_per_minute=1_000, tokens_per_minute=5_000_000),
@@ -190,7 +190,7 @@ _SPECS: tuple[ProviderSpec, ...] = (
impl=f"{_LLM}.kimi:KimiProvider",
label="Kimi",
default_model="kimi-for-coding",
- models=("kimi-for-coding", "kimi-for-coding-highspeed", "kimi-k2.5", "kimi-k2.6"),
+ models=("kimi-for-coding", "kimi-for-coding-highspeed", "kimi-k2.6"),
api_key_envs=("KIMI_API_KEY",),
base_url_envs=("KIMI_BASE_URL",),
rate_limit=RateLimitConfig(requests_per_minute=1_000, tokens_per_minute=5_000_000),
@@ -206,9 +206,9 @@ _SPECS: tuple[ProviderSpec, ...] = (
default_model="mistral/mistral-small-latest",
models=(
"mistral/mistral-small-latest",
- "openai/gpt-4o-mini",
- "anthropic/claude-haiku-4-5",
- "google/gemini-2.5-flash",
+ "openai/gpt-latest",
+ "anthropic/claude-sonnet-latest",
+ "vertex/gemini-flash-latest",
),
api_key_envs=("EDENAI_API_KEY",),
base_url_envs=("EDENAI_BASE_URL",),
@@ -240,8 +240,8 @@ _SPECS: tuple[ProviderSpec, ...] = (
picker_rank=11,
impl=f"{_LLM}.litellm:LiteLLMProvider",
label="LiteLLM",
- default_model="groq/llama-3.1-70b-versatile",
- models=("groq/llama-3.1-70b-versatile",),
+ default_model="groq/openai/gpt-oss-120b",
+ models=("groq/openai/gpt-oss-120b",),
api_key_envs=("LITELLM_API_KEY",),
base_url_envs=("LITELLM_BASE_URL", "LITELLM_API_BASE"),
keyless=True,
@@ -258,8 +258,8 @@ _SPECS: tuple[ProviderSpec, ...] = (
default_model="claude_cli/claude-haiku-5-5",
models=(
"claude_cli/claude-haiku-5-5",
- "claude_cli/claude-sonnet-4-6",
- "claude_cli/claude-opus-4-6",
+ "claude_cli/claude-sonnet-5-5",
+ "claude_cli/claude-opus-5-5",
),
keyless=True,
# Not needs_repo_cwd: it runs in a scratch dir so the repo's CLAUDE.md
@@ -290,7 +290,7 @@ _SPECS: tuple[ProviderSpec, ...] = (
impl=f"{_LLM}.opencode:OpenCodeProvider",
label="OpenCode (Local CLI)",
default_model="opencode/default",
- models=("opencode/default", "opencode/openai/gpt-5", "opencode/deepseek/deepseek-v4-pro"),
+ models=("opencode/default", "opencode/openai/gpt-6-luna", "opencode/deepseek/deepseek-v4-pro"),
keyless=True,
needs_repo_cwd=True,
local=True,
diff --git a/tests/unit/generation/test_cost_pricing.py b/tests/unit/generation/test_cost_pricing.py
index 77d216270..0c90ea7aa 100644
--- a/tests/unit/generation/test_cost_pricing.py
+++ b/tests/unit/generation/test_cost_pricing.py
@@ -6,11 +6,16 @@ the Sonnet-tier default rate, which overstates a cheap indexing run by ~40x.
from __future__ import annotations
+import pytest
+
+from repowise.core.cost_estimator.pricing import lookup_cost
from repowise.core.generation.cost_tracker import (
_FALLBACK_PRICING,
_family_pricing,
get_model_pricing,
+ resolve_model_pricing,
)
+from repowise.core.providers.llm.specs import PROVIDER_SPECS
def test_nano_models_priced_at_nano_rate_not_fallback() -> None:
@@ -37,6 +42,21 @@ def test_luna_priced_from_its_own_row_not_the_gpt5_flagship_tier() -> None:
assert get_model_pricing("gpt-5.6-luna") == {"input": 0.20, "output": 1.20}
+ "model",
+ [
+ model
+ for name in ("openai", "anthropic", "gemini", "deepseek", "openrouter")
+ for model in PROVIDER_SPECS[name].models
+ ],
+)
+def test_every_listed_paid_model_has_a_rate_in_both_tables(model: str) -> None:
+ """A picker model missing from a table estimates as free and records the fallback."""
+ rates = lookup_cost(model)
+ assert rates is not None and rates != (0.0, 0.0)
+ assert resolve_model_pricing(model) is not None
+
+
def test_unknown_model_still_falls_back() -> None:
assert get_model_pricing("totally-made-up-model-9000") == _FALLBACK_PRICING
diff --git a/tests/unit/test_providers/test_claude_cli_provider.py b/tests/unit/test_providers/test_claude_cli_provider.py
index 52b2ffdf2..21662aaea 100644
--- a/tests/unit/test_providers/test_claude_cli_provider.py
+++ b/tests/unit/test_providers/test_claude_cli_provider.py
@@ -145,8 +145,8 @@ def test_available_model_options_are_labelled(claude_on_path):
options = ClaudeCliProvider().available_model_options()
assert [o.model for o in options] == [
"claude_cli/claude-haiku-5-5",
- "claude_cli/claude-sonnet-4-6",
- "claude_cli/claude-opus-4-6",
+ "claude_cli/claude-sonnet-5-5",
+ "claude_cli/claude-opus-5-5",
]
assert sum(1 for o in options if o.recommended) == 1
assert all(
diff --git a/tests/unit/test_providers/test_cost_estimator.py b/tests/unit/test_providers/test_cost_estimator.py
index 8dbdaa4fd..382135b19 100644
--- a/tests/unit/test_providers/test_cost_estimator.py
+++ b/tests/unit/test_providers/test_cost_estimator.py
@@ -22,6 +22,8 @@ from repowise.cli.cost_estimator.heuristics import heuristic_tokens
# The default. An unpriced model is estimated as free (lookup_cost
# returns None), so this row is the guard.
("gpt-5.6-luna", 0.0002, 0.0012),
+ ("gpt-6-luna", 0.0001, 0.0005),
+ ("gpt-6-sol", 0.002, 0.010),
# Gemini
("gemini-3.1-flash-lite-preview", 0.00025, 0.0015),
("gemini-3-flash-preview", 0.0005, 0.003),
@@ -31,6 +33,8 @@ from repowise.cli.cost_estimator.heuristics import heuristic_tokens
("claude-sonnet-4-6", 0.003, 0.015),
("claude-haiku-4-5", 0.001, 0.005),
("claude-haiku-5-5", 0.0001, 0.0005),
+ ("claude-sonnet-5-5", 0.002, 0.01),
+ ("claude-opus-5-5", 0.004, 0.02),
# Free/local models
("mock", 0.0, 0.0),
("llama3", 0.0, 0.0),
diff --git a/tests/unit/test_providers/test_deepseek_provider.py b/tests/unit/test_providers/test_deepseek_provider.py
index d3797af7b..10240b0b8 100644
--- a/tests/unit/test_providers/test_deepseek_provider.py
+++ b/tests/unit/test_providers/test_deepseek_provider.py
@@ -26,7 +26,7 @@ def test_provider_name():
def test_default_model_is_flash():
p = DeepSeekProvider(api_key="sk-test")
- assert p.model_name == "deepseek-v4-flash"
+ assert p.model_name == "deepseek-flash"
def test_api_key_from_env(monkeypatch):
@@ -79,7 +79,7 @@ def test_available_model_options_uses_models_endpoint(monkeypatch):
assert captured["url"] == "https://api.deepseek.com/models"
assert captured["headers"] == {"Authorization": "Bearer sk-test"}
- flash = next(option for option in options if option.model == "deepseek-v4-flash")
+ flash = next(option for option in options if option.model == "deepseek-flash")
assert flash.reasoning_modes == (
"auto",
"off",
@@ -92,8 +92,8 @@ def test_available_model_options_uses_models_endpoint(monkeypatch):
"max",
)
assert flash.recommended is True
- current_flash = next(option for option in options if option.model == "deepseek-flash")
- assert current_flash.reasoning_modes == flash.reasoning_modes
+ legacy_flash = next(option for option in options if option.model == "deepseek-v4-flash")
+ assert legacy_flash.reasoning_modes == flash.reasoning_modes
def _make_mock_chat_response(
@@ -321,7 +321,7 @@ async def test_cost_tracker_called():
mock_tracker.record.assert_called_once()
call_kwargs = mock_tracker.record.call_args.kwargs
- assert call_kwargs["model"] == "deepseek-v4-flash"
+ assert call_kwargs["model"] == "deepseek-flash"
assert call_kwargs["input_tokens"] == 120
assert call_kwargs["output_tokens"] == 60
diff --git a/tests/unit/test_providers/test_openai_provider.py b/tests/unit/test_providers/test_openai_provider.py
index 916788f36..52cff7ace 100644
--- a/tests/unit/test_providers/test_openai_provider.py
+++ b/tests/unit/test_providers/test_openai_provider.py
@@ -69,11 +69,16 @@ def test_gpt56_family_ladder_drops_minimal_and_max():
exactly these five. `minimal` is gone from 5.6 entirely, and offering it
fails only on a real call, which is what this pins.
"""
- for model in ("gpt-5.6-luna", "gpt-5.6-sol"):
+ for model in ("gpt-5.6-luna", "gpt-5.6-sol", "gpt-6-luna", "gpt-6-sol"):
assert OpenAIProvider(
api_key="sk-test",
model=model,
).supported_reasoning_modes() == ("auto", "none", "low", "medium", "high", "xhigh")
+ # gpt-6.1 also drops `none` (verified live on gpt-6.1-sol, 2026-10-15).
+ assert OpenAIProvider(
+ api_key="sk-test",
+ model="gpt-6.1-sol",
+ ).supported_reasoning_modes() == ("auto", "low", "medium", "high", "xhigh")
# The generic gpt-5 branch is a prefix of gpt-5.6 and must not win.
assert (
@@ -308,7 +313,7 @@ async def test_generate_sends_correct_messages():
assert messages[1] == {"role": "user", "content": "user msg"}
[email protected]("model", ["gpt-5.6-luna", "gpt-5-mini", "o3", "o1", "o4-mini"])
[email protected]("model", ["gpt-6-luna", "gpt-5.6-luna", "gpt-5-mini", "o3", "o1", "o4-mini"])
async def test_generate_clamps_temperature_for_reasoning_models(model):
"""GPT-5+ / o-series models only accept the default temperature of 1."""
provider = OpenAIProvider(api_key="sk-test", model=model)
@@ -443,7 +448,7 @@ async def test_generate_rejects_minimal_for_non_reasoning_model():
@pytest.mark.parametrize(
"model",
- ["gpt-5.1", "gpt-5-pro", "gpt-5.6-luna", "gpt-5.6-sol"],
+ ["gpt-5.1", "gpt-5-pro", "gpt-5.6-luna", "gpt-5.6-sol", "gpt-6-luna", "gpt-6.1-sol"],
)
async def test_generate_rejects_minimal_for_known_unsupported_reasoning_models(model):
provider = OpenAIProvider(api_key="sk-test", model=model)
@@ -488,10 +493,12 @@ class _EmptyChatStream:
"gpt-5.6-terra",
"gpt-5.6-luna",
"openai/gpt-5.6-sol-2026-08-01",
+ "gpt-6-luna",
+ "gpt-6-sol",
],
)
async def test_stream_chat_disables_gpt_5_6_reasoning_with_function_tools(model):
- """Use the Chat Completions-compatible setting across the GPT-5.6 family."""
+ """Use the Chat Completions-compatible setting across the GPT-5.6 and GPT-6 families."""
provider = OpenAIProvider(api_key="sk-test", model=model)
captured_kwargs: list[dict] = []