benchr

Reference·June 2026

AI model deprecations: dates, replacements, and migration checks

The permanent record of AI model retirements: what's going dark, when, what replaces it, and what the swap does to your bill.

· DeepSeek's retirement notice and pricing page re-read on the provider docs, September 11, 2026 · View changelog

Claude modelsretired June 15, 2026
OpenAI modelsshut down Oct 23, 2026
Anthropic noticeminimum, public models
OpenAI noticeminimum, GA models

Current and upcoming shutdowns in 2026

Each model below has an announced end date from its provider. The migration column is the provider's official pick. Where benchr's pick differs (usually on price), the linked page explains why.

Announced retirements, re-verified August 31, 2026 against official provider documentation
ModelProviderShuts downOfficial replacementDetails
Imagen 4 API models (standard, Ultra, Fast)GoogleAug 17, 2026 (earliest — passed, removal unconfirmed)Gemini 3.1 Flash Imageimagen-4.0-generate-001, -ultra-, and -fast-; Google's table still lists the date without marking the rows shut down (checked Aug 31)
Gemini Robotics ER 1.6 PreviewGoogleAug 31, 2026Gemini Robotics ER 2 PreviewAnnounced with the ER 2 public preview on July 30, 2026
Sora 2 / Sora 2 ProOpenAISep 24, 2026None listedVideos API sunsets the same day; see the video roundup
Gemini 2.5 ProGoogleOct 16, 2026Gemini 3.1 Pro PreviewMigration page
Gemini 2.5 FlashGoogleOct 16, 2026Gemini 3.6 FlashMigration page; input price rises 5×
GPT-4oOpenAIOct 23, 2026GPT-5.5Migration page
GPT-4, GPT-4 Turbo, GPT-3.5OpenAIOct 23, 2026GPT-5.5 / GPT-5.4 miniOctober wave page
o1, o1-pro, o3-mini, o4-miniOpenAIOct 23, 2026GPT-5.5 / GPT-5.5 ProOctober wave page
GPT-4.1 nano, GPT Image 1OpenAIOct 23, 2026GPT-5.4 nano / GPT Image 2gpt-4.1 and 4.1-mini have no announced date
Prompts API · Evals · Agent BuilderOpenAINov 30, 2026Agents SDK (Agent Builder)Platform retirements, announced June 3, 2026
GPT Image 1 mini / 1.5OpenAIDec 1, 2026GPT Image 2Announced June 2, 2026
Older GPT-5 / o3 snapshotsOpenAIDec 11, 2026GPT-5.5 / 5.4 mini / 5.4 nano / 5.5 ProDated snapshots only (gpt-5-2025-08-07, o3-2025-04-16, …); floating aliases stay. Announced June 11, 2026
Legacy audio, realtime & transcription modelsOpenAIJan 20, 2027GPT Realtime 2.1 / GPT Audio 1.5Nine IDs, announced July 20, 2026: gpt-realtime, gpt-audio, gpt-4o-realtime, gpt-4o-audio, their mini variants, and gpt-4o-mini-transcribe-2025-03-20

The 2026 retirement timeline

  1. Mar 9Gemini 3 Pro Preview

    Shut down on the Gemini API after under four months. Vertex AI listed it discontinued by March 26.

  2. Apr 20Claude Haiku 3

    Retired. The last Claude 3-generation model to go.

  3. May 12DALL-E 2 and 3

    Removed from the API in favor of GPT Image 2.

  4. Jun 1Gemini 2.0 Flash family

    Shut down. Google's current recommended path is Gemini 3.6 Flash.

  5. Jun 15Claude Sonnet 4 + Opus 4

    Both retired on the Claude API. The Claude 4.x line replaces them at the same or lower price.

  6. Jun 25Gemini image previews

    Gemini 3 Pro Image Preview and 3.1 Flash Image Preview shut down; the stable gemini-3-pro-image and gemini-3.1-flash-image replace them.

  7. Jul 23OpenAI legacy IDs

    Five Codex IDs, both deep-research snapshots, computer-use-preview, and two chat snapshots retired on their recorded deadline.

  8. Jul 24DeepSeek aliases + Opus 4.7 fast mode

    DeepSeek's legacy aliases became inaccessible after 15:59 UTC; Opus 4.7 fast requests now return an error. Standard Opus 4.7 remains available.

  9. Aug 5Claude Opus 4.1

    Retired exactly one year after its August 5, 2025 release. Calls to claude-opus-4-1-20250805 now fail.

  10. Aug 10OpenAI chat snapshots + Google Gemini embedding preview

    Google now renders embedding-2-preview as shut down. OpenAI continues to show its two chat-snapshot records under the May 8 shutdown notice; the listed date has passed, but benchr does not state an independent runtime test.

  11. Aug 17Imagen 4 API models

    Google lists standard, Ultra, and Fast Imagen 4 API IDs with August 17 as the earliest possible shutdown date and gemini-3.1-flash-image as the replacement. The current rows are not yet rendered as shut down.

  12. Aug 31Kimi K2.5 + Moonshot V1

    Moonshot's current model list schedules full platform sunset for Kimi K2.5 and all Moonshot V1 text and vision endpoints. Kimi K3 is the current upgrade path.

  13. Sep 10DeepSeek-V4-Flash + Vision-Exp

    DeepSeek retired both hosted models the day it launched DeepSeek-V4.1-Flash. The old IDs route to the new model for now, at its lower rate.

  14. Oct 16Gemini 2.5 Pro + Flash

    Both model IDs share the same recorded retirement date. The listed replacements cost more in this comparison, so recalculate the workload before cutover.

  15. Oct 23OpenAI model-ID retirements

    The recorded date covers GPT-4o, GPT-4, GPT-4 Turbo, GPT-3.5 Turbo, GPT-4.1 nano, GPT Image 1, o1, o1-pro, o3-mini, and o4-mini IDs.

How each provider retires models

Anthropic labels models active, legacy, deprecated, then retired, and commits to at least 60 days' notice for publicly released models. Each active model carries a published "not sooner than" date: Opus 4.8, for instance, won't retire before May 28, 2027. Dates apply to the Claude API, AWS, and Microsoft Foundry; Bedrock and Vertex run their own schedules. Anthropic has also committed to preserving model weights and says it hopes to make past models available again someday.

OpenAI gives at least six months for generally available models, three for specialized variants, and as little as two weeks for previews. The April 22, 2026 announcement that ends the GPT-4 era was the largest single deprecation notice any provider has issued — and it didn't stop at models. A follow-up on June 3 scheduled three platform products for November 30: the Reusable Prompts API, the Evals dashboard and API, and Agent Builder (whose official path forward is the Agents SDK).

Google publishes earliest-possible shutdown dates rather than firm ones, then confirms exact dates closer in. Preview models can go fast. Gemini 3 Pro Preview lasted 16 weeks.

Moonshot AI publishes current availability in its model list. Following Kimi K3's launch, it closed Kimi K2.5 and Moonshot V1 endpoints to newly registered users and scheduled their full platform sunset for August 31, 2026.

DeepSeek has retired models with no notice period: V4-Flash and V4-Flash-Vision-Exp went the day V4.1-Flash launched, and their IDs now route to the replacement. The same September 10 note said V4-Pro traffic would move to V4.1-Flash on September 14; DeepSeek later withdrew that and says V4-Pro continues with billing unchanged. Treat a DeepSeek date as provisional until it passes.

Past recorded retirement dates

The permanent record: dates that have passed; provider runtime status can lag a published deadline
ModelProviderRecorded dateSuccessor
DeepSeek-V4-Flash + V4-Flash-Vision-ExpDeepSeekSep 10, 2026DeepSeek-V4.1-Flash
Kimi K2.5 + Moonshot V1 seriesMoonshot AIAug 31, 2026Kimi K3
Assistants APIOpenAIAug 26, 2026Responses + Conversations API
Prompt tools API + WorkbenchAnthropicAug 17, 2026No successor — prompt tooling moves into your code
GPT-5.2 / GPT-5.3 chat-latestOpenAIAug 10, 2026GPT-5.6 Sol
Gemini Embedding 2 PreviewGoogleAug 10, 2026Gemini Embedding 2
Claude Opus 4.1AnthropicAug 5, 2026Claude Opus 4.8
o3 / o4-mini-deep-researchOpenAIJul 23, 2026GPT-5.5 Pro
gpt-5 / 5.1-chat-latestOpenAIJul 23, 2026GPT-5.5
Codex models (gpt-5-codex → 5.2-codex)OpenAIJul 23, 2026GPT-5.5 / GPT-5.4 mini
computer-use-previewOpenAIJul 23, 2026GPT-5.4 mini
deepseek-chat / deepseek-reasoner (legacy aliases)DeepSeekJul 24, 2026DeepSeek V4-Flash / V4-Pro
Claude Opus 4.7 fast modeAnthropicJul 24, 2026Claude Opus 4.8 fast mode
Gemini 3 Pro Image PreviewGoogleJun 25, 2026Gemini 3 Pro Image (stable)
Gemini 3.1 Flash Image PreviewGoogleJun 25, 2026Gemini 3.1 Flash Image (stable)
Claude Sonnet 4AnthropicJun 15, 2026Claude Sonnet 4.6
Claude Opus 4AnthropicJun 15, 2026Claude Opus 4.8
Gemini 2.0 Flash familyGoogleJun 1, 2026Gemini 3.6 Flash
DALL-E 2 / DALL-E 3OpenAIMay 12, 2026GPT Image 2
Claude Haiku 3AnthropicApr 20, 2026Claude Haiku 4.5
Gemini 3 Pro PreviewGoogleMar 9, 2026Gemini 3.1 Pro Preview
Claude Sonnet 3.7AnthropicFeb 19, 2026Claude Sonnet 4.6
Claude Haiku 3.5AnthropicFeb 19, 2026Claude Haiku 4.5
Claude 3 OpusAnthropicJan 5, 2026Claude Opus 4.8
Claude Sonnet 3.5 (both snapshots)AnthropicOct 28, 2025Claude Sonnet 4.6

A 15-minute model-ID audit

Older model IDs often remain in production configuration after a newer default ships. Run these three checks:

1. Grep your codebase for every model string in the tables above, including date-suffixed snapshots like gpt-4o-2024-05-13 and claude-sonnet-4-20250514. Aliases hide in config files and environment variables, not just code.

Already seeing failures? A retired ID surfaces as model_not_found on OpenAI and not_found_error on Anthropic — the error database has the per-provider diagnosis and code fixes.

2. Pull your usage export. Anthropic's console has a per-key, per-model CSV export. OpenAI's usage dashboard breaks down by model. Five minutes tells you which deprecated IDs still get real traffic.

3. Price the replacement before you swap. Same-provider upgrades aren't always neutral. Gemini 2.5 Flash users face a 5× input price jump, while Claude Opus 4 users save two-thirds. Run your token volumes through the cost calculator before picking the target, and check the current rankings in case a different provider now fits better.

Frequently asked

Which AI models shut down next?

Google lists its three Imagen 4 API IDs for an August 17 earliest shutdown date, with Gemini 3.1 Flash Image as the provider-listed replacement. Google says it will communicate the exact date in advance. The next later deadline in this record is Assistants API on August 26.

What happens to API calls after retirement?

They fail. Each provider defines a lifecycle (active, deprecated, retired) and publishes a recommended replacement. Once a model passes its retirement date, requests to that ID return errors, and fine-tuned variants built on it stop working too.

How much notice do providers give?

Anthropic: at least 60 days for public models. OpenAI: at least 6 months for GA models, 3 months for specialized variants, about two weeks for previews. Google: publishes earliest-possible dates and confirms the real one later.

Do replacement models cost more?

Sometimes less, sometimes a lot more. Opus 4 to Opus 4.8 cuts your bill by two-thirds. Gemini 2.5 Flash to 3.6 Flash multiplies input cost by five. Every migration page here shows the before-and-after math so you can decide with numbers.

Changelog

  • — Recorded DeepSeek's September 10 retirement of the hosted V4-Flash and V4-Flash-Vision-Exp models, with DeepSeek-V4.1-Flash as the official replacement, and noted that DeepSeek withdrew the V4-Pro phase-out it had announced for September 14.
  • — Moonshot's docs confirm Kimi K2.5 and the Moonshot V1 series retired today; calls now return a 404, and the record moves to the permanent table with Kimi K3 as the listed replacement. The Assistants API record also moved to the permanent table after its August 26 date passed and OpenAI filed the platform under past deprecations. Imagen 4's August 17 earliest date has passed without Google marking the rows shut down, so that record stays deprecated. The lifecycle dataset now holds 48 records.
  • — Both August 10 deadlines have passed and were re-read at the source: OpenAI now files gpt-5.2-chat-latest and gpt-5.3-chat-latest under past deprecations, and Google's embedding-2-preview date is behind us, so both moved into the permanent record. Added four lifecycle records the dataset did not carry: OpenAI's July 20 deprecation of nine legacy audio, realtime, and transcription IDs (removal January 20, 2027), Google's August 17 Imagen 4.0 shutdown, Google's August 31 shutdown of gemini-robotics-er-1.6-preview, and Anthropic's August 17 retirement of the experimental prompt tools API and the Workbench. The Assistants API deadline on August 26 is now the closest one on the page.
  • — Added Google's embedding-2-preview deadline and official gemini-embedding-2 replacement to the permanent dataset. Re-verified OpenAI's two chat-snapshot deadlines. Both remain marked deprecated because the official tables still present them as scheduled records on the date itself.
  • — Claude Opus 4.1 moved from "shutdowns ahead" to "retired": its August 5, 2026 shutdown has passed, re-confirmed on Anthropic's official model-deprecations table. Moved the row into the permanent-record table and logged the flip in deprecations.json.
  • — Moved the July 23 OpenAI records into the past-date table and marked the July 24 records as today's deadlines. The page does not claim runtime removal without a fresh provider or API-status confirmation.
  • — Refreshed the current-deadline summary for the OpenAI legacy-ID wave recorded for July 23 and the Claude/DeepSeek records dated July 24. Historical retirements remain in the table.
  • — Gemini 3 Pro Image Preview and Gemini 3.1 Flash Image Preview moved from "shutdowns ahead" to "retired": their June 25, 2026 shutdown has passed, re-confirmed on Google's official deprecation table. Replacements are the GA gemini-3-pro-image and gemini-3.1-flash-image. Logged in deprecations.json.
  • — Added OpenAI's June 11, 2026 deprecation of older GPT-5 and o3 dated snapshots (gpt-5-2025-08-07, gpt-5-mini/nano-2025-08-07, gpt-5-pro-2025-10-06, o3-2025-04-16, o3-pro-2025-06-10), removed from the API December 11, 2026. The floating gpt-5 / gpt-5-mini / gpt-5-nano aliases stay live. Verified on the official OpenAI deprecations page; logged in deprecations.json.
  • — Claude Sonnet 4 and Claude Opus 4 moved from "shutdowns ahead" to "retired": their June 15, 2026 shutdown has passed, re-confirmed on Anthropic's deprecation table. Opus 4.1 remains upcoming (August 5, 2026).
  • — Same-day expansion after a full re-read of all three providers' deprecation docs: added OpenAI's July 23 wave in full (five Codex IDs, o4-mini-deep-research, computer-use-preview), GPT-4.1 nano and GPT Image 1 to the October 23 list, the November 30 platform retirements (Prompts API, Evals, Agent Builder — announced June 3), and Gemini 3 Pro Image Preview (June 25). Logged Anthropic's official "not sooner than" retirement floors for all active Claude models in the dataset.
  • — Hub published with 14 upcoming retirements and 8 historical entries, all verified against Anthropic, OpenAI, and Google official deprecation docs.
  • — Google now renders embedding-2-preview as shut down, so the permanent record moved it to retired. Added all three Imagen 4 API IDs with their August 17 earliest shutdown date and gemini-3.1-flash-image replacement. OpenAI's GPT-5.2 / GPT-5.3 snapshots remain deprecated: the August 10 listed date has passed, but the provider has not marked the record as past.
  • — Added Google's embedding-2-preview deadline and official gemini-embedding-2 replacement to the permanent dataset. Re-verified OpenAI's two chat-snapshot deadlines. Both remain marked deprecated because the official tables still present them as scheduled records on the date itself.
  • — Claude Opus 4.1 moved from "shutdowns ahead" to "retired": its August 5, 2026 shutdown has passed, re-confirmed on Anthropic's official model-deprecations table. Moved the row into the permanent-record table and logged the flip in deprecations.json.
  • — Moved the July 23 OpenAI records into the past-date table and marked the July 24 records as today's deadlines. The page does not claim runtime removal without a fresh provider or API-status confirmation.
  • — Refreshed the current-deadline summary for the OpenAI legacy-ID wave recorded for July 23 and the Claude/DeepSeek records dated July 24. Historical retirements remain in the table.
  • — Gemini 3 Pro Image Preview and Gemini 3.1 Flash Image Preview moved from "shutdowns ahead" to "retired": their June 25, 2026 shutdown has passed, re-confirmed on Google's official deprecation table. Replacements are the GA gemini-3-pro-image and gemini-3.1-flash-image. Logged in deprecations.json.
  • — Added OpenAI's June 11, 2026 deprecation of older GPT-5 and o3 dated snapshots (gpt-5-2025-08-07, gpt-5-mini/nano-2025-08-07, gpt-5-pro-2025-10-06, o3-2025-04-16, o3-pro-2025-06-10), removed from the API December 11, 2026. The floating gpt-5 / gpt-5-mini / gpt-5-nano aliases stay live. Verified on the official OpenAI deprecations page; logged in deprecations.json.
  • — Claude Sonnet 4 and Claude Opus 4 moved from "shutdowns ahead" to "retired": their June 15, 2026 shutdown has passed, re-confirmed on Anthropic's deprecation table. Opus 4.1 remains upcoming (August 5, 2026).
  • — Same-day expansion after a full re-read of all three providers' deprecation docs: added OpenAI's July 23 wave in full (five Codex IDs, o4-mini-deep-research, computer-use-preview), GPT-4.1 nano and GPT Image 1 to the October 23 list, the November 30 platform retirements (Prompts API, Evals, Agent Builder — announced June 3), and Gemini 3 Pro Image Preview (June 25). Logged Anthropic's official "not sooner than" retirement floors for all active Claude models in the dataset.
  • — Hub published with 14 upcoming retirements and 8 historical entries, all verified against Anthropic, OpenAI, and Google official deprecation docs.

Sources

  • Anthropic model deprecations — platform.claude.com/docs/en/about-claude/model-deprecations (verified June 12, 2026)
  • OpenAI deprecations — developers.openai.com/api/docs/deprecations (verified August 21, 2026)
  • Gemini API deprecations — ai.google.dev/gemini-api/docs/deprecations (verified August 21, 2026)
  • Moonshot AI model list — platform.kimi.ai/docs/models (verified August 21, 2026)
  • Anthropic deprecation commitments — anthropic.com/research/deprecation-commitments
  • benchr deprecations dataset (open JSON) — every entry carries its source and verification date
Related