Q4 2026 AI Model Deprecation Calendar: Every Sunset Date Across OpenAI, DashScope, Anthropic, and Google

Q4 2026 is the quarter when the bill comes due for every team that postponed model migrations. OpenAI's October 23 sweep alone removes GPT-3.5 Turbo, GPT-4, GPT-4 Turbo, o1, o1-pro, o3-mini, o4-mini, gpt-image-1, and every fine-tuned model built on those bases (Deprecations, retrieved 2026-09-21). DashScope retires another batch of legacy mainline and snapshot models on October 10 across six separate notices (Model Depreciation, retrieved 2026-09-21). Anthropic's earliest "not sooner than" floors start arriving: Claude Sonnet 4.5 on September 29 and Claude Haiku 4.5 on October 15 (Model deprecations, retrieved 2026-09-21). Google is retiring gemini-2.5-flash-image on October 2, with broader Gemini 2.5 retirements still unscheduled (AI Models Discontinued in 2026, retrieved 2026-09-21).

This page is the unified calendar. We built it because no single provider publishes a cross-provider deprecation view, and if you route requests across OpenAI, Anthropic, DashScope, and Google, you need one place to see what's coming and when to update your routing config.

We route OpenAI-compatible requests through configured providers and support provider/model routing and fallback where the live product path supports it. We do not claim universal model support, zero downtime, or guaranteed cheapest pricing. Every date and model ID below comes from official vendor documentation at the retrieval time listed.

OpenAI-compatible means a provider exposes a chat-completions endpoint whose request and response shape matches the OpenAI API contract closely enough that an unmodified OpenAI SDK call works against it after swapping three values: API key, base URL, and model name. The minimum surface in practice is POST /v1/chat/completions with messages, model, and an OpenAI-shaped streaming response.

30-second answer

• October 1, 2026: OpenAI retires gpt-5.4-cyber (replacement: gpt-5.6-cyber).
• October 2, 2026: Google retires gemini-2.5-flash-image (replacement: gemini-3.1-flash-image-preview).
• October 10, 2026: DashScope sunsets legacy mainline and snapshot models across 6 separate notices, covering old Qwen series and third-party hosted models.
• October 23, 2026: OpenAI's largest single-day retirement: GPT-3.5 Turbo, GPT-4, GPT-4 Turbo, o1, o1-pro, o3-mini, o4-mini, gpt-image-1, gpt-4.1-nano, and all fine-tuned models on those bases.
• October 31, 2026: OpenAI Evals platform becomes read-only.
• November 24, 2026: Anthropic's "not sooner than" floor for Claude Opus 4.5 (claude-opus-4-5-20251101).
• November 30, 2026: OpenAI shuts down reusable prompts API, Evals dashboard, and Agent Builder.
• December 1, 2026: OpenAI retires gpt-image-1-mini, gpt-image-1.5, and chatgpt-image-latest (replacement: gpt-image-2).
• December 11, 2026: OpenAI retires GPT-5 and o3/o3-pro snapshots (replacement: GPT-5.6 Sol/Terra/Luna).
• If you use a routing layer with automatic fallback, you can pre-configure replacement models and test before each sunset hits.

Complete Q4 2026 timeline

All dates below are sorted chronologically. Each entry links to the official deprecation notice.

September 24, 2026 (late Q3 overlap)

September 28, 2026 (late Q3 overlap)

October 1, 2026

October 2, 2026

October 10, 2026

DashScope retires legacy models across 6 notices. This covers both mainline models and snapshot models on Alibaba Cloud Bailian.

For detailed DashScope model-by-model migration paths, see our dedicated guide: Oct 2026 DashScope Massive Model Sunset Migration Guide.

October 15, 2026 (Anthropic floor)

This is a "not sooner than" floor, not a confirmed retirement date. Anthropic guarantees the model remains available until at least this date. No deprecation has been announced.

October 23, 2026

The largest single-day API retirement in 2026.

October 31, 2026

November 24, 2026 (Anthropic floor)

November 30, 2026

December 1, 2026

December 11, 2026

Anthropic lifecycle floors in Q4 2026

Anthropic does not publish hard retirement dates for active models. Instead, it publishes "not sooner than" floors, guaranteeing each model remains available until at least the listed date. Here are the floors that fall within Q4 2026:

Anthropic provides at least 60 days' notice before any retirement and has publicly committed to preserving retired model weights (Model deprecations, retrieved 2026-09-21).

Google Gemini lifecycle in Q4 2026

Google's deprecation approach differs from other providers. Published shutdown dates represent floors ("earliest possible dates"), not deadlines. Google communicates exact dates separately, typically with at least six months of warning after general availability.

Some third-party deprecation trackers list October 16, 2026 as a firm retirement date for Gemini 2.5 Pro and Flash. This appears to originate from notices sent to Gemini Enterprise Agent Platform customers, which is a different product surface from the Gemini API. Google's own Gemini API deprecations page lists these models with no shutdown date.

Cross-provider migration matrix

This table maps each retiring model to the recommended replacement. Use it to update your routing config before each date.

Routing fallback configuration

If you use a routing layer with automatic model fallback, you can prepare for each sunset date by adding replacement models to your fallback chain. Test the replacement model on your workloads before the sunset date, then let the fallback handle the transition.

For teams using TheRouter, the model fallback configuration lets you define ordered fallback chains per model. When a model returns a 404 (which is what happens after a retirement date), the router automatically tries the next model in the chain. We support this for OpenAI-compatible providers where routing and fallback are implemented in the live product path.

Example: pre-configuring for October 23 retirements:

{
  "model": "gpt-4-turbo",
  "fallbacks": ["gpt-5.6-sol"]
}

{
  "model": "o4-mini",
  "fallbacks": ["gpt-5.6-terra"]
}

For DashScope models, see TheRouter's model fallback documentation for provider-specific configuration.

Decision tree: planning your Q4 migrations

1. Audit your model usage now. Check which retiring models your applications call. OpenAI provides a usage export in the Platform Dashboard. Anthropic offers a similar export in Claude Console. DashScope usage is available in the Bailian Console.

2. Prioritize by date. October 10 (DashScope) and October 23 (OpenAI) are the two largest waves. If you use both providers, plan DashScope migrations first since they hit 13 days earlier.

3. Test replacements on your actual workloads. The recommended replacement models are generally more capable than the retiring ones, but they may behave differently on your specific prompts, output formats, and latency requirements.

4. Update routing config. If you use a routing layer, add fallback chains before each sunset date. Test the fallback path before you need it.

5. Watch Anthropic's announcements. The "not sooner than" floors mean Anthropic could announce a deprecation at any time after the floor date. Monitor the Model deprecations page and your email for notices.

6. Watch Google's Gemini 2.5 status. No retirement date is published, but when Gemini 3 reaches general availability, Google has said it will announce retirement with at least six months of warning.

FAQ

How many models retire on October 23?

OpenAI retires at least 12 model IDs plus their fine-tuned variants on October 23, 2026, making it the single largest API model retirement day of the year (Deprecations, retrieved 2026-09-21).

Does Anthropic have any hard retirement dates in Q4 2026?

No. Anthropic publishes "not sooner than" floors, not hard retirement dates. The nearest floors in Q4 are Claude Sonnet 4.5 (September 29), Claude Haiku 4.5 (October 15), and Claude Opus 4.5 (November 24). No deprecation has been announced for any of these models yet (Model deprecations, retrieved 2026-09-21).

Is Gemini 2.5 Pro being retired in October?

No confirmed retirement date exists. Some third-party trackers list October 16, 2026, but this appears to originate from Gemini Enterprise Agent Platform notices, not the Gemini API. Google's official Gemini API deprecation page lists the model with "No shutdown date announced" (AI Models Discontinued in 2026, retrieved 2026-09-21).

What happens if I call a retired model after the shutdown date?

For OpenAI, the API returns an error (typically HTTP 404). For DashScope, the API stops returning results and existing applications using the model will fail. For Anthropic, requests to retired models fail with an error response.

Can I get extended access to a model past its retirement date?

OpenAI offers dedicated capacity for continued access to some retiring models through their sales team. Anthropic has committed to preserving retired model weights but does not currently offer extended API access. DashScope does not offer extensions past shutdown dates.

What is the replacement for GPT-3.5 Turbo?

OpenAI recommends gpt-5.6-terra as the replacement for GPT-3.5 Turbo. This is a significant capability upgrade, so test on your workloads to verify output format compatibility and cost impact (Deprecations, retrieved 2026-09-21).

Will fine-tuned GPT-4 models stop working on October 23?

Yes. Fine-tuned models built on GPT-4 and GPT-3.5 Turbo bases will also be retired on October 23, 2026. OpenAI recommends migrating to fine-tuning on the GPT-5.6 family if you still need fine-tuning (Deprecations, retrieved 2026-09-21).

Related content

• Oct 2026 DashScope Massive Model Sunset Migration Guide
• OpenAI Oct 2026 Model Deprecation Planning Guide
• Chinese AI Model Deprecation Calendar Q3 2026
• Anthropic Model Deprecation Cadence Migration Guide
• OpenAI GPT-5 Deprecation Migration Planning Guide
• DashScope Model Lifecycle Consolidation Reference
• LLM API Fallback Routing Guide
• OpenAI provider page
• DashScope provider page
• Anthropic provider page
• Google provider page