πͺ’ Open source agent evals & observability: Trace, evaluate, and improve LLM applications with one open platform.
TypeScript analyticsautogenevaluationlangchainlarge-language-modelsllama-indexllmllm-evaluationllm-observabilityllmopsmonitoringobservabilityopen-sourceopenaiplaygroundprompt-engineeringprompt-managementself-hostedycombinator
chore(pricing): add Gemini 4 Argon pricing and selectable-model coverage (#18762)
langfuse-botf8a29ac37485
diff --git a/.agents/skills/add-model-price/references/model-audit-memory.md b/.agents/skills/add-model-price/references/model-audit-memory.mdindex 545960d8fe..63b1e710c8 100644--- a/.agents/skills/add-model-price/references/model-audit-memory.md+++ b/.agents/skills/add-model-price/references/model-audit-memory.md@@ -12,12 +12,74 @@ audit date.## Latest useful snapshot-**Audit date:** 2026-10-13+**Audit date:** 2026-10-14All prices listed as `$X / MTok` (per million tokens). Per-token JSON values: divide by 1,000,000.-The 2026-10-13 run re-fetched the full Anthropic pricing page (model table, cache-hits-footnote, Fast mode, Batch, and tool-use tables) and the Anthropic model-deprecations+The 2026-10-14 run re-fetched the full Anthropic pricing page (model table, cache-hits+footnote, Fast mode, Batch, and tool-use tables) and the Anthropic model-deprecations page,+the OpenAI aggregate Standard pricing table and the OpenAI deprecations page, both Gemini+3.x pricing fetches and the Gemini models catalog, the AWS Bedrock pricing page, and the+TypeSafe Jev models page. Every existing price matched verbatim; no price drift was found.+This run found one new model and two lifecycle findings, none requiring a selectable-array+removal yet:+1. **Gemini 4 Argon** (`gemini-4-argon`) β a new Google flagship reasoning model, confirmed+ via `ai.google.dev/gemini-api/docs/models`, its dedicated model page, and the pricing+ page. "New Stable" status, 1,048,576-token unified input/output context window, max+ output 1,048,576 tokens with long decode (65,536 default synchronous). Supports+ text/image/video/audio/PDF input, text-only output, thinking (low/medium/high), function+ calling, Search grounding, grounding with Google Maps, and implicit-only prompt caching+ (no explicit cache-write support). Standard tier priced at introductory rates through+ November 30, 2026: $2.00/MTok input, $10.00/MTok output (including thinking tokens),+ $0.10/MTok cache read (10% ratio, matching the universal Gemini cache-read ratio) β+ stepping up to $4.00/$20.00/$0.20 on December 1, 2026 (a different, earlier step-up date+ than the 3.6/3.7/3.8 Flash family's January 1, 2027 date β do not conflate the two). No+ context-length pricing threshold (flat price across the full 1M window, unlike+ `gemini-3.1-pro-preview`/`gemini-2.5-pro`'s <=200K/>200K split). Grounding is $14 per+ 1,000 requests (5,000 free/month, shared across Gemini 3-and-newer models), same rate as+ the rest of the Gemini 3.x family. A Priority tier (`service_tier: "priority"`) is+ separately quoted on the pricing page at exactly 1.8x the Standard rate at every+ introductory/stepped-up price point ($3.60/$18.00/$0.18 through Nov 30, 2026, doubling+ with the same step-up on Dec 1, 2026) β same multiplier as every other Gemini model with+ a Priority tier. Not available on the Free tier. Added to the pricing file (mirroring the+ `gemini-3.8-flash`/`gemini-3.6-flash` key set, Standard + Priority tiers) and to+ `googleAIStudioModels` only (not `vertexAIModels`): the Vertex AI lifecycle page+ (`cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versions`) 301-redirects to+ `docs.cloud.google.com`, outside the allowed WebFetch domain list, so Vertex-side+ availability for this specific new model is unconfirmed β report as a standing gap if a+ future audit needs to add it to `vertexAIModels` once that domain is reachable or+ independently confirmed. matchPattern: `(?i)^(google(ai)?/)?(gemini-4-argon)$`, verified+ via the bundled match-pattern tester to accept `gemini-4-argon`/`googleai/gemini-4-argon`/+ `google/gemini-4-argon` and reject `gemini-4-argon-preview`/`gemini-3.8-flash`/a dated+ snapshot guess. No dated snapshot was found at launch, consistent with other recent+ flagship launches.+2. **New OpenAI deprecation announced the same day (2026-10-14): a "GPT-4 series" batch,+ shutdown January 14, 2027** β confirmed via two independent verbatim-quote fetches of+ `developers.openai.com/api/docs/deprecations`. Exact rows: GPT-4o+ (`gpt-4o-2024-08-06`, `gpt-4o-2024-11-20`) to `gpt-6-luna`; GPT-4o mini+ (`gpt-4o-mini-2024-07-18`) to `gpt-6-luna`; GPT-4.1 (`gpt-4.1-2025-04-14`) to+ `gpt-6-luna`; GPT-4.1 mini (`gpt-4.1-mini-2025-04-14`) to `gpt-6-luna`. Note+ `gpt-4o-2024-11-20` is a dated snapshot not currently present in `openAIModels` (only+ `gpt-4o`, `gpt-4o-2024-08-06`, and `gpt-4o-2024-05-13` are), so it has no+ selectable-array action regardless. January 14, 2027 is three months after this audit's+ date β not yet actionable under the removal criteria. Re-check+ `developers.openai.com/api/docs/deprecations` on or after that date and remove+ `gpt-4o`/`gpt-4o-2024-08-06`/`gpt-4.1`/`gpt-4.1-2025-04-14`/`gpt-4.1-mini`/+ `gpt-4.1-mini-2025-04-14` from `openAIModels` once the page confirms the shutdown has+ passed (keep pricing entries). This is a separate, later batch from the already-tracked+ October 23, 2026 "Legacy GPT model snapshots" batch below β do not conflate the two+ shutdown dates or model lists.+3. **Standing re-checks, both still not yet passed**: the October 23, 2026 "Legacy GPT model+ snapshots" batch (`gpt-3.5-turbo-0125`/`gpt-4-0613`/`gpt-4-1106-preview`/`gpt-4-turbo`/+ `gpt-4.1-nano`/`gpt-4o-2024-05-13`/`o1`/`o1-pro`/`o3-mini`/`o4-mini`, see the 2026-10-09+ entry below) is still upcoming (9 days away) β re-confirmed via the deprecations page's+ own "Upcoming deprecations" section heading. Claude Haiku 4.5's tentative retirement+ ("Not sooner than October 15, 2026") is also still one day in the future and still has no+ firm announced date β re-confirmed via the Anthropic model-deprecations page's "Model+ status" table (`claude-haiku-4-5-20251001` still `Active`, `Deprecated: N/A`). Neither+ triggers a removal yet.+The 2026-10-13 run before it re-fetched the full Anthropic pricing page (model table,+cache-hits footnote, Fast mode, Batch, and tool-use tables) and the Anthropic model-deprecationspage, targeted rows of the OpenAI aggregate Standard pricing table (the full gpt-6/gpt-5.6family plus every gpt-5.x/gpt-4.x/o-series/legacy model and `gpt-5-chat-latest`'s owndedicated page), the OpenAI deprecations page and model catalog, the Gemini 3.x and 2.5@@ -211,7 +273,8 @@ to have siblings `gpt-5.5-cyber`/`gpt-5.4-cyber`, see provider-sources-and-price| Google | gemini-3-pro-preview | Input $2/$4 (β€200k/>200k), Output $12/$18 | No | Large Context (>200K) set in file | Not applicable | None | https://ai.google.dev/pricing https://ai.google.dev/gemini-api/docs/models | Still not listed on either official page this run; existing prices retained, not re-verified. || Google | gemini-3.6-flash | Input $0.75, Output $3.75, Cache read $0.075 (all through Dec 31, 2026; reverts to $1.50/$7.50/$0.15 from Jan 1, 2027) | Yes | No large-context tier | Yes | None | https://ai.google.dev/pricing https://ai.google.dev/gemini-api/docs/pricing | Re-confirmed still on introductory pricing; wording and step-up date unchanged. Now the "previous-generation" Flash model behind gemini-3.7-flash and gemini-3.8-flash. || Google | gemini-3.7-flash | Same introductory pricing/step-up date as gemini-3.6-flash | Yes | No large-context tier | Yes | None | https://ai.google.dev/pricing https://ai.google.dev/gemini-api/docs/models | Re-confirmed unchanged; description changed from "New Stable" to "previous-generation" now that gemini-3.8-flash is GA. |-| Google | gemini-3.8-flash | Input $0.75, Output $3.75, Cache read $0.075 (through Dec 31, 2026; steps to $1.50/$7.50/$0.15 Jan 1, 2027) | Yes | No large-context tier | Yes | Added | https://ai.google.dev/pricing https://ai.google.dev/gemini-api/docs/pricing https://ai.google.dev/gemini-api/docs/models | New "New Stable" GA model, direct successor to gemini-3.7-flash, described as engineered "for long-horizon software engineering, autonomous agents, and complex enterprise workflows." Same intro pricing and Jan 1 2027 step-up as 3.6/3.7-flash. Added mirroring the gemini-3.7-flash key set exactly, and to `vertexAIModels`/`googleAIStudioModels` (not first entry). |+| Google | gemini-3.8-flash | Input $0.75, Output $3.75, Cache read $0.075 (through Dec 31, 2026; steps to $1.50/$7.50/$0.15 Jan 1, 2027) | Yes | No large-context tier | Yes | None | https://ai.google.dev/pricing https://ai.google.dev/gemini-api/docs/pricing https://ai.google.dev/gemini-api/docs/models | Re-confirmed unchanged. |+| Google | gemini-4-argon | Standard: Input $2.00, Output $10.00 (incl. thinking), Cache read $0.10 (through Nov 30, 2026; steps to $4.00/$20.00/$0.20 Dec 1, 2026). Priority (`service_tier: "priority"`, 1.8x Standard at every date): $3.60/$18.00/$0.18 through Nov 30, 2026, doubling with Standard on Dec 1, 2026 | Yes | No context-length pricing threshold (flat 1,048,576-token window); Standard + Priority tiers only | Yes | Added | https://ai.google.dev/gemini-api/docs/models https://ai.google.dev/gemini-api/docs/models/gemini-4-argon https://ai.google.dev/gemini-api/docs/pricing | New flagship reasoning model, "New Stable" status. 1M unified input/output context, max output 1,048,576 with long decode (65,536 default). Text/image/video/audio/PDF input, text-only output, thinking (low/medium/high) billed at the output rate, function calling, Search + Maps grounding ($14/1,000 queries, same shared quota as other Gemini 3.x-and-newer models), implicit-only caching (no explicit cache-write key). Different step-up date (Dec 1, 2026) than the 3.6/3.7/3.8-flash family's Jan 1, 2027 date β do not conflate. Added to `googleAIStudioModels` only; `vertexAIModels` left unresolved because the Vertex AI lifecycle page redirects outside the allowed WebFetch domain list (see the standing quirk note in provider-sources-and-price-keys.md). || Google | gemini-2.0-flash / gemini-2.0-flash-001 | Input $0.10, Output $0.40 | No | Deprecated (shut down June 1, 2026) | Not applicable | None | https://ai.google.dev/pricing | Not re-verified this run; retained for backward compatibility. |## Unresolved findings (updated 2026-09-14)diff --git a/.agents/skills/add-model-price/references/provider-sources-and-price-keys.md b/.agents/skills/add-model-price/references/provider-sources-and-price-keys.mdindex c0ff4a86ca..9ff4bac769 100644--- a/.agents/skills/add-model-price/references/provider-sources-and-price-keys.md+++ b/.agents/skills/add-model-price/references/provider-sources-and-price-keys.md@@ -1223,6 +1223,71 @@ json_literal = "<price_per_mtok>e-6"were left unchanged this run, carried forward from their last confirmed values($1.50/$9.00/$0.15 and $0.75/$3.75/$0.075 respectively).+- **Gemini 4 Argon (added October 14 2026)** β `gemini-4-argon` is Google's new+ flagship reasoning model, confirmed "New Stable" via+ `https://ai.google.dev/gemini-api/docs/models`, its dedicated page+ `https://ai.google.dev/gemini-api/docs/models/gemini-4-argon`, and+ `https://ai.google.dev/gemini-api/docs/pricing`. Unified 1,048,576-token+ input/output context window; max output 65,536 tokens by default, up to+ 1,048,576 with long decode. Input: text, image, video, audio, PDF. Output:+ text only. Supports thinking (low/medium/high, billed at the output rate β+ no separate reasoning price), function calling, Search grounding, grounding+ with Google Maps, and implicit-only prompt caching (explicit caching is+ **not** supported β do not add a cache-write key family). Standard pricing+ is introductory through November 30, 2026: $2.00/MTok input, $10.00/MTok+ output (including thinking tokens), $0.10/MTok cache read (the universal+ 10% Gemini cache-read ratio) β stepping up to $4.00/$20.00/$0.20 on+ **December 1, 2026**. This is a different step-up date from the+ 3.6/3.7/3.8-flash family's January 1, 2027 date; do not conflate the two+ when revisiting either family after their respective dates. No+ context-length pricing threshold β flat pricing across the full window,+ unlike `gemini-3.1-pro-preview`/`gemini-2.5-pro`'s <=200K/>200K split.+ Grounding is $14 per 1,000 requests (5,000 free/month, shared across+ Gemini-3-and-newer models), the same rate as the rest of the Gemini 3.x+ family. A Priority tier (`service_tier: "priority"`) is separately+ verbatim-quoted on the pricing page at exactly 1.8x the Standard rate at+ every pricing point ($3.60/$18.00/$0.18 through Nov 30, 2026, doubling in+ lockstep with Standard on Dec 1, 2026) β the same multiplier documented for+ every other Gemini model with a Priority tier (see the September 14 2026+ Priority finding above). Not available on the Free tier. Added to the+ pricing file mirroring the `gemini-3.8-flash`/`gemini-3.6-flash` Standard ++ Priority key set exactly, and to `googleAIStudioModels` only (not the first+ entry). **Not added to `vertexAIModels`**: the Vertex AI lifecycle page+ (`cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versions`)+ 301-redirects to `docs.cloud.google.com`, outside the allowed WebFetch+ domain list (the same standing quirk documented in "Official Model+ Lifecycle Sources" above) β re-check Vertex availability once that domain+ is reachable or independently confirmed. matchPattern:+ `(?i)^(google(ai)?\/)?(gemini-4-argon)$`, verified via the bundled+ match-pattern tester against `gemini-4-argon`, `googleai/gemini-4-argon`,+ `google/gemini-4-argon` (accept) and `gemini-4-argon-preview`,+ `gemini-3.8-flash`, a guessed dated snapshot (reject). No dated snapshot+ was found at launch.+- **New OpenAI "GPT-4 series" deprecation batch announced October 14 2026,+ shutdown January 14, 2027 (not yet actionable)** β Two independent+ verbatim-quote fetches of `https://developers.openai.com/api/docs/deprecations`+ confirm this exact table:++ | Shutdown date | Model / system | Recommended replacement |+ | ---------------- | --------------------------------------------------- | ------------------------ |+ | January 14, 2027 | GPT-4o (`gpt-4o-2024-08-06`, `gpt-4o-2024-11-20`) | `gpt-6-luna` |+ | January 14, 2027 | GPT-4o mini (`gpt-4o-mini-2024-07-18`) | `gpt-6-luna` |+ | January 14, 2027 | GPT-4.1 (`gpt-4.1-2025-04-14`) | `gpt-6-luna` |+ | January 14, 2027 | GPT-4.1 mini (`gpt-4.1-mini-2025-04-14`) | `gpt-6-luna` |++ `gpt-4o-2024-11-20` is a dated snapshot not currently present in+ `openAIModels` (only `gpt-4o`, `gpt-4o-2024-08-06`, and `gpt-4o-2024-05-13`+ are), so it has no selectable-array action regardless of date. January 14,+ 2027 is three months after this finding's date β not yet actionable under+ the removal criteria in `automated-audit.md`. Re-check this page on or+ after that date and remove `gpt-4o`, `gpt-4o-2024-08-06`, `gpt-4.1`,+ `gpt-4.1-2025-04-14`, `gpt-4.1-mini`, and `gpt-4.1-mini-2025-04-14` from+ `openAIModels` once the page confirms the shutdown has passed (keep their+ pricing entries). This is a **separate, later batch** from the+ already-tracked October 23, 2026 "Legacy GPT model snapshots" batch (see+ the October 9 2026 entry above) β do not conflate the two shutdown dates or+ model lists when a future audit processes either one.+## Provider Usage KeysUse [provider-usage-key-matrix.md](provider-usage-key-matrix.md) as the singlediff --git a/packages/shared/src/server/llm/types.ts b/packages/shared/src/server/llm/types.tsindex ccd645f325..0a88d3d140 100644--- a/packages/shared/src/server/llm/types.ts+++ b/packages/shared/src/server/llm/types.ts@@ -443,6 +443,7 @@ export const vertexAIModels = [export const googleAIStudioModels = ["gemini-2.5-flash","gemini-2.5-pro",+ "gemini-4-argon","gemini-3.8-flash","gemini-3.7-flash","gemini-3.6-flash",diff --git a/worker/src/constants/default-model-prices.json b/worker/src/constants/default-model-prices.jsonindex a3036905be..16009b8f0e 100644--- a/worker/src/constants/default-model-prices.json+++ b/worker/src/constants/default-model-prices.json@@ -8439,6 +8439,82 @@}]},+ {+ "id": "f076cdab-4e50-4edb-9135-b5065c3cd0d6",+ "modelName": "gemini-4-argon",+ "matchPattern": "(?i)^(google(ai)?\/)?(gemini-4-argon)$",+ "createdAt": "2026-10-14T00:00:00.000Z",+ "updatedAt": "2026-10-14T00:00:00.000Z",+ "pricingTiers": [+ {+ "id": "f076cdab-4e50-4edb-9135-b5065c3cd0d6_tier_default",+ "name": "Standard",+ "isDefault": true,+ "priority": 0,+ "conditions": [],+ "prices": {+ "input": 2e-6,+ "input_text": 2e-6,+ "input_modality_1": 2e-6,+ "prompt_token_count": 2e-6,+ "promptTokenCount": 2e-6,+ "input_cached_tokens": 0.1e-6,+ "input_cache_read": 0.1e-6,+ "cached_content_token_count": 0.1e-6,+ "output": 10e-6,+ "output_text": 10e-6,+ "output_modality_1": 10e-6,+ "candidates_token_count": 10e-6,+ "candidatesTokenCount": 10e-6,+ "thoughtsTokenCount": 10e-6,+ "thoughts_token_count": 10e-6,+ "output_reasoning_tokens": 10e-6,+ "output_reasoning": 10e-6,+ "grounding_queries": 14e-3,+ "groundingQueries": 14e-3,+ "web_search_queries": 14e-3,+ "webSearchQueries": 14e-3+ }+ },+ {+ "id": "62843157-f0a0-4e10-ad86-5dd985033bd6",+ "name": "Priority",+ "isDefault": false,+ "priority": 1,+ "conditions": [+ {+ "source": "model_parameters",+ "key": "service_tier",+ "operator": "in",+ "values": ["priority"]+ }+ ],+ "prices": {+ "input": 3.6e-6,+ "input_text": 3.6e-6,+ "input_modality_1": 3.6e-6,+ "prompt_token_count": 3.6e-6,+ "promptTokenCount": 3.6e-6,+ "input_cached_tokens": 0.18e-6,+ "input_cache_read": 0.18e-6,+ "cached_content_token_count": 0.18e-6,+ "output": 18e-6,+ "output_text": 18e-6,+ "output_modality_1": 18e-6,+ "candidates_token_count": 18e-6,+ "candidatesTokenCount": 18e-6,+ "thoughtsTokenCount": 18e-6,+ "thoughts_token_count": 18e-6,+ "output_reasoning_tokens": 18e-6,+ "output_reasoning": 18e-6,+ "grounding_queries": 14e-3,+ "groundingQueries": 14e-3,+ "web_search_queries": 14e-3,+ "webSearchQueries": 14e-3+ }+ }+ ]+ },{"id": "b2c3d4e5-f6a7-4b8c-9d0e-1f2a3b4c5d6e","modelName": "gemini-3.5-flash-lite",