improvement(providers): audit and update all provider model definitions

[waleedlatif1]

Summary

• Added missing models: gpt-5.2-pro, gpt-5-pro, o3-mini, o3-pro, claude-sonnet-4-6, gemini-3.1-flash-lite-preview, deepseek-reasoner, grok-4.20 models, mistral-small-2603, devstral-2512, devstral-latest
• Fixed Anthropic context windows and max output tokens verified against live API
• Fixed OpenAI gpt-4.1 family context (1047576) and max output tokens (32768)
• Updated DeepSeek, xAI, Mistral pricing to match official docs
• Fixed Cerebras context windows to exact values
• Removed shut down/deprecated models (gemini-3-pro-preview, 5 Groq models, 2 Cerebras models, 2 retired Mistral models)
• Extended adaptive thinking support to Sonnet 4.6 and Opus 4.5
• Updated all updatedAt dates to 2026-04-01

Type of Change

• Enhancement (non-breaking change that improves existing functionality)

Testing

Tested manually

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
docs	Skipped		Apr 2, 2026 2:14am

[cursor]

PR Summary

Medium Risk
Medium risk because it changes the provider model catalog (added/removed models) and updates token limits/pricing used for request sizing and cost calculation, plus adjusts Anthropic thinking-mode selection for Sonnet 4.6 vs other Claude variants.

Overview
Updates the provider model registry across OpenAI/Azure, Anthropic, Google/Vertex, DeepSeek, xAI, Cerebras, Groq, Mistral, and Bedrock: adds new model IDs, removes deprecated entries, and refreshes most pricing.updatedAt values.

Corrects several model capability fields that drive runtime behavior—especially contextWindow and maxOutputTokens (notably OpenAI gpt-4.1* and multiple Claude models), plus assorted pricing/cached-input adjustments.

Refines Anthropic extended-thinking configuration so thinking.type: "adaptive" is only used for claude-opus-4-6 and claude-sonnet-4-6 (others use budget_tokens), and updates provider utils tests to match the new model set and revised token limits.

^{Written by Cursor Bugbot for commit c518c6f. Configure here.}

[greptile-apps]

Greptile Summary

This PR performs a broad audit and update of all provider model definitions across OpenAI, Anthropic, Google Gemini/Vertex, DeepSeek, xAI, Groq, Cerebras, and Mistral. It adds ~15 new models, removes deprecated/shut-down models, corrects context windows and max-output-token limits against live API responses, and updates all updatedAt timestamps to 2026-04-01.

Key changes:

• Anthropic: Added claude-sonnet-4-6 with 1M context and 128K max output; extended adaptive thinking (type: "adaptive") to cover both Opus 4.6 and Sonnet 4.6; corrected claude-opus-4-1/claude-opus-4-0 max output from 64 K → 32 K; fixed context windows from 200 K → 1 M for several Sonnet/Opus 4.x models
• OpenAI: Added gpt-5.2-pro, gpt-5-pro, o3-mini, o3-pro; corrected gpt-4.1 family context from 1 M → 1,047,576 and added 32 K max-output cap
• Google: Replaced deprecated gemini-3-pro-preview with gemini-3.1-flash-lite-preview (both direct and Vertex); added minimal thinking level across Gemini 3 models; added cached-input pricing for gemini-2.0-flash
• DeepSeek: Updated pricing for all three models; added deepseek-reasoner — however its pricing matches the V3 chat model ($0.28/$0.028/$0.42) rather than the R1 reasoning model ($0.55/$0.14/$2.19), which may misrepresent costs
• xAI: Added three grok-4.20-0309 variants with 2M context; corrected cached-input pricing for several Grok 3/4 models
• Groq: Removed five deprecated models
• Cerebras: Removed two deprecated models; corrected context windows from approximate (131 000) to exact (131 072) values
• Mistral: Added mistral-small-2603, devstral-2512, devstral-latest; removed two retired models; updated pricing for mistral-small-latest and corrected codestral-latest context from 256 K → 128 K

Confidence Score: 4/5

• Safe to merge after verifying deepseek-reasoner pricing; one P1 data inconsistency and one P2 test gap remain.
• The adaptive-thinking fix from the previous thread is correctly addressed. The bulk of the changes (date updates, context window corrections, model additions/removals) are routine and well-structured. The one P1 concern is that deepseek-reasoner is assigned V3 chat pricing instead of R1 reasoning pricing, which would silently under-report costs for any user calling that model ID. The P2 gap is a missing test assertion for claude-sonnet-4-6 thinking support. Neither issue causes a crash or security problem, but the pricing inaccuracy is a real data correctness defect on the changed path.
• apps/sim/providers/models.ts — verify deepseek-reasoner pricing against official DeepSeek API docs

Important Files Changed

Filename	Overview
apps/sim/providers/models.ts	Large-scale model registry update: new models added (gpt-5.2-pro, gpt-5-pro, o3-mini, o3-pro, claude-sonnet-4-6, gemini-3.1-flash-lite-preview, deepseek-reasoner, grok-4.20 variants, Mistral additions), deprecated models removed, pricing/context/max-output-token corrections applied across all providers. One pricing inconsistency found: deepseek-reasoner uses V3 chat pricing instead of R1 pricing.
apps/sim/providers/anthropic/core.ts	Extended supportsAdaptiveThinking() to match sonnet-4-6 strings alongside existing opus-4-6 patterns; comment updated to document which models support adaptive vs budget-tokens thinking. Clean, well-commented change.
apps/sim/providers/utils.test.ts	Test suite updated to remove gemini-3-pro-preview assertions, add gpt-5.2-pro/gpt-5-pro reasoning-effort assertions, update claude-opus-4-1 max-output expectation from 64000 to 32000; missing assertion for new claude-sonnet-4-6 thinking support.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[API call with modelId] --> B{supportsAdaptiveThinking?}
    B -- opus-4-6 or sonnet-4-6 --> C[thinking type adaptive\nplus output_config]
    B -- other thinking model --> D[thinking type enabled\nplus budget_tokens]
    B -- non-thinking model --> E[No thinking config]
    C --> F[Anthropic API]
    D --> F
    E --> F

    subgraph "Notable new models in PR"
        H[claude-sonnet-4-6\nadaptive thinking]
        I[deepseek-reasoner\npricing needs verify]
        J[gpt-5.2-pro and gpt-5-pro\nreasoning effort only]
        K[o3-mini and o3-pro]
        L[gemini-3.1-flash-lite-preview\nreplaces gemini-3-pro-preview]
    end

Loading

_{Reviews (3): Last reviewed commit: "fix(providers): move maxOutputTokens ins..." | Re-trigger Greptile}

[waleedlatif1]

@greptile

[waleedlatif1]

@cursor review

[waleedlatif1]

@greptile

[waleedlatif1]

@cursor review

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR

[waleedlatif1]

@greptile

[waleedlatif1]

@cursor review

[cursor]

✅ Bugbot reviewed your changes and found no new issues!

Comment @cursor review or bugbot run to trigger another review on this PR