Repository navigation
improvement(models): retire deep research, refresh model lifecycle, pricing, and reseller catalogs - #10255
Conversation
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
There was a problem hiding this comment.
All reported issues were addressed across 36 files
Reply with feedback, questions, or to request a fix.
Turn on auto-fix | Re-trigger cubic
[High impact] The PR appears safe to merge; no new blocking issue was found. SummaryRetires Gemini Deep Research, refreshes model availability and defaults, and updates reseller catalogs and prices.
|
|
@cubic-dev-ai review this PR |
@waleedlatif1 I have started the AI code review. It will take a few minutes to complete. |
|
@cubic-dev-ai review this PR |
@waleedlatif1 I have started the AI code review. It will take a few minutes to complete. |
…long-context tiers
…the catalog pages
… Completions providers
…keep meter types local
3a3a557 to
92a5ebe
Compare
Summary
deep-research-pro-preview-12-2025on Oct 23, 2026, and it was the only model on the Interactions API. Drops the request/poll/stream path ingemini/core.ts, thedeepResearchcapability and helpers,previousInteractionId/interactionIdplumbing in the agent handler, block and provider types, and the landing/modelsdeep-research facets. The agent block'spreviousInteractionIdfield goes throughSUBBLOCK_ID_MIGRATIONS(_removed_), and the bundled templates drop it. Both catalog entries stay asdeprecatedso pinned workflows still validatedeprecated(OpenAI o1/o3-mini/o4-mini, Vertex Gemini 2.5 and 3.6 Flash, Azure o3/o4-mini/gpt-5-chat, Azure Anthropic Opus 4.1, DeepSeek chat/reasoner, 13 retired Mistral ids, Groq Llama 3.x and Qwen 3.6, Cerebras GLM 4.7 and Gemma 4, four retired NVIDIA NIM models, Nebius Qwen3 235B) and deprecated-but-callable or aliased oneslegacy; reseller copies of superseded Claude models mirror the directlegacystatusdefaultModelnow points at a live model, using the replacement each provider names (e.g. Vertex/Google → Gemini 3.8 Flash, Groq → gpt-oss-120b, Bedrock → Sonnet 5.5)gpt-6.1-terramistral-large-latest(now Large 4), Codestral 128K context, Z.ai cached-input prices, Groq Qwen 3.8 / KIE gpt-6.1-sol context windows, Muse Spark 1.3maxeffort, Bedrock Claude 5.xforcedToolUse/nativeStructuredOutputsmirroring the direct entriescachedInputrate for Z.ai, Kimi, DeepSeek, and Groq (they reported cache hits insideprompt_tokensand were billed at the full input rate). A sharedcreateChatCompletionUsageMeterprices each call's uncached remainder and cache read, also per call so a long-context tier applies per request; xAI and the shared streaming tool loop use it tooType of Change
Testing
providers/openai-compat/cache-billing.test.ts, plususage-meter.test.ts)bunx vitest runover providers/utils, providers/models, providers/bedrock, providers/gemini, landing models, agent handler, workflow builders, real-catalog (524 passed);bun run generateclean;check:agent-stream-docs;check-block-registry.ts origin/staging(subblock ID stability passes with the migration)Checklist
test-auditauthoring gate)