chore(cognitive): update AI model catalog

[github-actions]

Model Update Summary

Updated 38 models across 7 providers (openai: no changes, anthropic, google-ai, groq, cerebras, xai, openrouter, fireworks-ai)

---

Anthropic

defaultModel: claude-sonnet-4-5-20250929 → claude-sonnet-4-6

New Models Added

Model ID	Change	Notes
claude-opus-4-6	NEW	Production; $5/$25/1M tokens; 1M context / 128k output
claude-sonnet-4-6	NEW	Production; $3/$15/1M tokens; 1M context / 64k output
claude-opus-4-5-20251101	NEW	Production; $5/$25/1M tokens; 200k context / 64k output
claude-opus-4-1-20250805	NEW	Production; $15/$75/1M tokens; 200k context / 32k output
claude-opus-4-20250514	NEW	Production; $15/$75/1M tokens; 200k context / 32k output

Updated Models

Model ID	Field	Old → New
claude-3-haiku-20240307	lifecycle	production → deprecated
claude-3-haiku-20240307	deprecationDate	(none) → 2026-04-19
claude-3-haiku-20240307	replacementModels	(none) → ['claude-haiku-4-5-20251001']

---

Google AI

New Models Added

Model ID	Change	Notes
gemini-3.1-pro	NEW	Preview; internalModelId gemini-3.1-pro-preview; $2/$12/1M tokens; 1M context / 65k output
gemini-3.1-flash-lite	NEW	Preview; internalModelId gemini-3.1-flash-lite-preview; $0.25/$1.50/1M tokens; 1M context / 65k output
gemini-2.5-flash-lite	NEW	Production (stable); $0.10/$0.40/1M tokens; 1M context / 65k output

Updated Models

Model ID	Field	Old → New
gemini-3-pro	lifecycle	preview → discontinued
gemini-3-pro	discontinuedDate	(none) → 2026-03-09 (shut down by Google)
gemini-3-pro	replacementModels	(none) → ['gemini-3.1-pro']
gemini-3-pro	tags	['reasoning', ...] → ['deprecated', 'reasoning', ...]
gemini-2.0-flash	lifecycle	production → deprecated
gemini-2.0-flash	deprecationDate	(none) → 2026-01-01
gemini-2.0-flash	replacementModels	(none) → ['gemini-2.5-flash']
gemini-2.0-flash	tags	['low-cost', ...] → ['deprecated', 'low-cost', ...]

---

Groq

New Models Added

Model ID	Change	Notes
llama-4-scout-17b-16e-instruct	NEW	Production/Preview; internalModelId meta-llama/llama-4-scout-17b-16e-instruct; $0.11/$0.34/1M tokens; vision support
kimi-k2-instruct-0905	NEW	Production/Preview; internalModelId moonshotai/kimi-k2-instruct-0905; $1.00/$3.00/1M tokens
qwen3-32b	NEW	Production/Preview; internalModelId qwen/qwen3-32b; $0.29/$0.59/1M tokens

Updated Models

Model ID	Field	Old → New
gpt-oss-20b	inputCostPer1mTokens	0.1 → 0.075
gpt-oss-20b	outputCostPer1mTokens	0.5 → 0.3
gpt-oss-20b	maxOutputTokens	32_000 → 65_536
gpt-oss-120b	outputCostPer1mTokens	0.75 → 0.6
gpt-oss-120b	maxOutputTokens	32_000 → 65_536
llama-3.3-70b-versatile	maxInputTokens	128_000 → 131_000
llama-3.1-8b-instant	maxInputTokens	128_000 → 131_000
llama-3.1-8b-instant	maxOutputTokens	8192 → 131_072

---

Cerebras

New Models Added

Model ID	Change	Notes
qwen-3-235b-a22b-instruct-2507	NEW	Preview; Qwen3 235B MoE; $0.80/$1.60/1M tokens (est.); 131k context
zai-glm-4.7	NEW	Preview; Z.ai GLM 4.7 355B; $0.39/$1.75/1M tokens; 131k context

No other changes

---

xAI

defaultModel: grok-4-fast-non-reasoning → grok-4.20-0309-non-reasoning

New Models Added

Model ID	Change	Notes
grok-4.20-0309-reasoning	NEW	Production; $2/$6/1M tokens; 2M context / 128k output
grok-4.20-0309-non-reasoning	NEW	Production; $2/$6/1M tokens; 2M context / 128k output
grok-4.20-multi-agent-0309	NEW	Production; $2/$6/1M tokens; 2M context / 128k output
grok-4-1-fast-reasoning	NEW	Production; $0.20/$0.50/1M tokens; 2M context / 128k output
grok-4-1-fast-non-reasoning	NEW	Production; $0.20/$0.50/1M tokens; 2M context / 128k output

No existing model changes

---

OpenRouter

New Models Added

Model ID	Change	Notes
gpt-oss-20b	NEW	Production; $0.075/$0.30/1M tokens; 131k context / 65k output

Updated Models

Model ID	Field	Old → New
gpt-oss-120b	outputCostPer1mTokens	0.75 → 0.6
gpt-oss-120b	maxOutputTokens	32_000 → 65_536

---

Fireworks AI

New Models Added

Model ID	Change	Notes
deepseek-v3p2	NEW	Production; DeepSeek V3.2; $0.56/$1.68/1M tokens; 163k context
deepseek-v3p1	NEW	Production; DeepSeek V3.1; $0.56/$1.68/1M tokens; 163k context
kimi-k2-instruct-0905	NEW	Production; Kimi K2 0905; $0.60/$2.50/1M tokens; 262k context

Updated Models

Model ID	Field	Old → New
deepseek-v3-0324	inputCostPer1mTokens	0.9 → 0.56
deepseek-v3-0324	outputCostPer1mTokens	0.9 → 1.68
deepseek-v3-0324	maxInputTokens	160_000 → 163_840

---

OpenAI

OpenAI's documentation page (platform.openai.com/docs/models) returned 403 Forbidden and could not be fetched programmatically. No changes were made to the OpenAI config. Manual verification recommended.

---

Notes

• Cerebras pricing for new preview models (qwen-3-235b-a22b-instruct-2507, zai-glm-4.7) is estimated based on comparable model sizes from other providers, as Cerebras does not publish pricing publicly (requires dedicated endpoint agreement).
• gemini-3-pro was already shut down by Google as of March 9, 2026; updated to discontinued.
• gemini-2.0-flash is deprecated upstream by Google; updated lifecycle accordingly.
• claude-3-haiku-20240307 is deprecated by Anthropic with retirement date April 19, 2026.
• xAI introduced a new model naming scheme (grok-4.20-*) alongside the older grok-4-* naming; both coexist.