Model Updates

This page lists every change to a model in the Gloo AI catalog: models added, models deprecated or retired, and calls moved to a replacement model. It is a log of events. The rules behind them are on Model Lifecycle, and the models you can use today, with live prices, are on Supported Models.

​

How to read this

Each month has a table of changes, newest first, and one note per change under it, in the same order.

• Date is the day a change took effect in the API. A date marked “announced” is the period the Changelog announced it in, used where the change shipped over several days.
• Change is one of:
  • Added: the model joined the catalog.
  • Deprecated: the model still works and has a retirement date.
  • Redirected: calls to the model ID go to its replacement, and successful responses carry redirect headers.
  • Retired: nothing serves the model ID any more. From its retirement date it returns 410 with model_retired, or, for models removed before that error existed, it is no longer in the catalog. See When nothing replaces a model.
• Model → Replacement names the model, then the model your calls now go to. “none (410)” means calls return 410 with model_retired; “none (removed)” means the model left the catalog; “none” alone means no replacement was named. An added model is marked “(new)”; a batch of more than five shows only a count.
• The notes under each table give the model IDs (old → new), compare list prices as stated when the change was made, and say when a replacement behaves differently. For current prices, see Supported Models.

A retired model can drop out of GET /platform/v2/models before its retirement date, and its deprecation_note goes with it. This page keeps the record. Calls are billed at the rates of the model that serves them. How a replacement is picked, and when its price can be higher, is in How a replacement is chosen.

​

November 2026

• GPT-5 Mini (gloo-openai-gpt-5-mini → gloo-openai-gpt-6-luna): lower price. OpenAI retires GPT-5 Mini on December 11. GPT-6 Luna has no minimal effort, so minimal runs as low and is listed in ignored_params; none now turns reasoning off.
• GPT-5 Nano (gloo-openai-gpt-5-nano → gloo-openai-gpt-6-luna): about 2× the input price and 1.25× the output price, and more for prompts over 272,000 tokens. OpenAI retires GPT-5 Nano on December 11. Same effort changes as GPT-5 Mini.
• Kimi K2.5 (gloo-kimi-k2.5 → gloo-kimi-k2.6): higher price: as of November 2026, $0.77 / $3.56 per million input / output tokens, up from $0.57 / $2.85 (about 35% more per input token and 25% more per output token). Double the context window (262,144 tokens) and image input. Kimi K2.5 retires on November 18.
• MiniMax M2.7 (gloo-minimax-m2.7 → gloo-minimax-m3): same price, plus image input. Smaller listed context window: 524,288 tokens, against 1,000,000. MiniMax M2.7 retired on November 5.
• Mercury 3 (gloo-inception-mercury-3): returns 410 with model_retired from November 11. It retired on November 8, and until November 11 its calls went to GLM-5.3 Flash; no newer Mercury model exists. Choose another model from Supported Models.
• Mercury 2.5 (gloo-inception-mercury-2.5-preview): same as Mercury 3, returning 410 from November 11.
• GPT-5 Image Mini (gloo-openai-gpt-5-image-mini): returns 410 from December 1; until then, responses carry Deprecation and Sunset headers. To generate images, use the image_generation tool on a GPT-5.x model, or gloo-openai-gpt-5-image at a higher price.
• Gemini 2.5 Flash (gloo-google-gemini-2.5-flash → gloo-google-gemini-3.5-flash-lite): same price. Gemini 3.5 Flash Lite always reasons, so reasoning_effort: none no longer turns reasoning off. Google retires Gemini 2.5 Flash on December 1.
• Gemini 2.5 Pro (gloo-google-gemini-2.5-pro → gloo-google-gemini-3.8-flash): lower price. Both models always reason. Google retires Gemini 2.5 Pro on December 1.
• 5 models added on Nov 4–5:
  • GPT-6.1 Terra (gloo-openai-gpt-6.1-terra) for everyday coding and agent work, and GPT-6.1 Terra Pro (gloo-openai-gpt-6.1-terra-pro) for harder tasks; both take images and about a million tokens.
  • Kimi K3.1 Pro (gloo-kimi-k3.1-pro) takes images and a million tokens, and reasoning_effort: none turns its reasoning off.
  • Reflection Beam 1.0 (gloo-reflection-beam-1.0-501b-a23b) is a million-token reasoning model for coding and long-running agents.
  • LFM2.5-24B-A2B (gloo-liquid-lfm-2.5-24b-a2b) is a low-cost text model.

​

October 2026

• gpt-image-1 in the image_generation tool (gpt-image-1 → gpt-image-2): OpenAI retired gpt-image-1. A tool call that names it runs on gpt-image-2 and carries X-Gloo-Model-Redirected-From, Deprecation and Sunset headers.
• gpt-image-1.5, chatgpt-image-latest and gpt-image-1-mini in the image_generation tool (gpt-image-1.5, chatgpt-image-latest, gpt-image-1-mini): no replacement named. They still work and retire on December 1, 2026. Set the tool’s model to gpt-image-2, gpt-image-2.5-flare or gpt-image-2.5-sunburst.
• gpt-image-2, gpt-image-2.5-flare and gpt-image-2.5-sunburst in the image_generation tool (gpt-image-2, gpt-image-2.5-flare, gpt-image-2.5-sunburst): choose them with the tool’s model setting on most OpenAI GPT-5 and GPT-6 models. gpt-image-2 is used when you don’t set one. See Image generation tool.
• 7 models added on Oct 21:
  • Gemini 4 Argon (gloo-google-gemini-4-argon) is Google’s flagship for complex coding, math and science, and multi-step agents; it always reasons, at low, medium or high effort.
  • Meta Muse Spark Open (gloo-meta-muse-spark-open), NVIDIA Nemotron 3.5 Super VL (gloo-nvidia-nemotron-3.5-super-vl, images and video) and Moondream 3 Preview (gloo-moondream-3-preview) cover coding and vision.
  • DeepSeek V4.1 Flash 1017 (gloo-deepseek-v4.1-flash-1017), Mercury 3 (gloo-inception-mercury-3) and Pareto 26.10 (gloo-unbiased-pareto-26.10) take a million tokens. Mercury 3 retired on November 8.
• Ling 3.0 Flash Fin (gloo-inclusionai-ling-3.0-flash-fin): retired with no successor and no longer listed. Choose another model from Supported Models.
• Poolside Laguna S 2.1 (gloo-poolside-laguna-s-2.1 → gloo-z-ai-glm-5.3-flash): a model from another maker, chosen before replacements came from the same maker. Same million-token context, tools and reasoning, at a lower input price and a higher output price. Poolside Laguna S 2.1 retired on October 31 and is no longer listed.
• MiMo V2.5 (gloo-xiaomi-mimo-v2.5 → gloo-xiaomi-mimo-v2.6-flash): same price. MiMo V2.5 retired on October 21.
• MiMo V2.5 Pro (gloo-xiaomi-mimo-v2.5-pro → gloo-xiaomi-mimo-v2.6-pro): same price. MiMo V2.5 Pro retired on October 21.
• Qwen 3 235B (gloo-qwen-3-235b → gloo-qwen-3.7-plus): lower price and a larger context window. Retired on October 9.
• Qwen 3 Max (gloo-qwen-3-max → gloo-qwen-3.6-27b): lower price. Retired on October 9.
• Qwen 3.6 Max Preview (gloo-qwen-3.6-max-preview → gloo-mistral-large-4): Mistral Large 4, a model from another maker, chosen before replacements came from the same maker. Lower price and a larger context window. Retired on October 9.
• Qwen3 Coder Plus (gloo-qwen-3-coder-plus → gloo-qwen-3.6-plus): lower price. Retired on October 9.
• 32 models added between August and October. The catalog passed 160 models from 14 provider families, including embedding models, all behind one API key:
  • GPT-6 Astra, Sol and Luna with Pro variants (gloo-openai-gpt-6-astra, gloo-openai-gpt-6-astra-pro, gloo-openai-gpt-6-sol, gloo-openai-gpt-6-sol-pro, gloo-openai-gpt-6-luna, gloo-openai-gpt-6-luna-pro)
  • GPT-6.1 Sol and Sol Pro (gloo-openai-gpt-6.1-sol, gloo-openai-gpt-6.1-sol-pro)
  • GPT-5.6 Pro variants (gloo-openai-gpt-5.6-luna-pro, gloo-openai-gpt-5.6-sol-pro, gloo-openai-gpt-5.6-terra-pro)
  • Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 (gloo-anthropic-claude-opus-5.5, gloo-anthropic-claude-sonnet-5.5, gloo-anthropic-claude-haiku-5.5)
  • Grok 4.7 (gloo-xai-grok-4.7) and GLM-5.3 (gloo-z-ai-glm-5.3)
  • MiMo V2.6 (gloo-xiaomi-mimo-v2.6-flash, gloo-xiaomi-mimo-v2.6-pro)
  • Four NVIDIA Nemotron models (gloo-nvidia-nemotron-3.5-lightning, gloo-nvidia-nemotron-3-super-120b-a12b, gloo-nvidia-nemotron-3-ultra-550b-a55b, gloo-nvidia-nemotron-3-nano-30b-a3b)
  • Mistral Large 4 (gloo-mistral-large-4) and DeepSeek V4.1 Flash (gloo-deepseek-v4.1-flash)
  • The Qwen 3.8 family (gloo-qwen-3.8-27b, gloo-qwen-3.8-flash, gloo-qwen-3.8-omni-flash, gloo-qwen-3.8-max, gloo-qwen-3.8-max-prime, gloo-qwen-3.8-2.4t-a95b)
  • Command A+ (gloo-cohere-command-a-plus) and Gemini Nano Banana 2.1 (gloo-google-gemini-nano-banana-2.1)

​

September 2026

• Claude Sonnet 4.5 (gloo-anthropic-claude-sonnet-4.5 → gloo-anthropic-claude-sonnet-5): lower price. Claude Sonnet 5 always reasons. Anthropic retires Claude Sonnet 4.5 on November 30. On the Anthropic Messages endpoint, Claude Sonnet 4.5 returns 404 with model_retired, naming Claude Sonnet 5, instead of redirecting.
• GPT-4.1 Nano (gloo-openai-gpt-4.1-nano → gloo-google-gemini-2.5-flash-lite): Gemini 2.5 Flash Lite, a model from another maker, chosen before replacements came from the same maker. Same price. OpenAI retired GPT-4.1 Nano on October 23.
• DeepSeek V3.2 (gloo-deepseek-v3.2 → gloo-deepseek-v4-flash): lower price and a larger context window. DeepSeek V3.2 retired on September 28.
• Mistral Large 3 (gloo-mistral-large-3 → gloo-mistral-small-4): lower price. Mistral withdrew Mistral Large 3.
• Gemini 2.5 Flash Image (gloo-google-gemini-2.5-flash-image → gloo-google-gemini-3.1-flash-lite-image): equal or lower price for input, output and images, and a larger context window. Google retired Gemini 2.5 Flash Image on October 2.

​

August 2026

• DeepSeek R1 (gloo-deepseek-r1 → gloo-deepseek-r1-0528): lower price and a larger context window.
• 6 image models: FLUX.2 Klein 4B (gloo-bfl-flux-2-klein), FLUX.2 Pro (gloo-bfl-flux-2-pro), FLUX.2 Max (gloo-bfl-flux-2-max), FLUX.2 Flex (gloo-bfl-flux-2-flex), Seedream 4.5 (gloo-bytedance-seedream-4.5) and Grok Imagine (gloo-xai-grok-imagine). No longer listed from August 3 and removed on August 13. Generate images with a current image model from Supported Models or the image_generation tool.

​

July 2026

• Devstral 2 (gloo-mistral-devstral-2512 → gloo-qwen-3-coder-next): Qwen3 Coder Next 80B, a model from another maker, chosen before replacements came from the same maker. Lower price. Mistral withdrew Devstral 2.
• Claude Opus 4.1 (gloo-anthropic-claude-opus-4.1 → gloo-anthropic-claude-opus-5): lower price and a larger context window. Calls went to Claude Opus 4.8 until August 6.
• Claude Sonnet 4 (gloo-anthropic-claude-sonnet-4 → gloo-anthropic-claude-sonnet-4.6): same price. Deprecated on June 16.
• Claude Opus 4 (gloo-anthropic-claude-opus-4 → gloo-anthropic-claude-opus-4.8): larger context window. Deprecated on June 17.

​

June 2026

• MiMo V2 Flash (gloo-xiaomi-mimo-v2-flash): choose another model from Supported Models, such as MiMo V2.6 Flash.
• Claude Opus 4.6 Fast, 4.7 Fast and 4.8 Fast (gloo-anthropic-claude-opus-4.6-fast, gloo-anthropic-claude-opus-4.7-fast, gloo-anthropic-claude-opus-4.8-fast): use the standard Claude Opus models.

​

May 2026

• DeepSeek V3.2 Speciale (gloo-deepseek-v3.2-speciale): choose another DeepSeek model from Supported Models.
• 11 models added, week of May 26. The Opus Fast variants and MiMo V2 Flash were retired in June 2026.
  • Claude Opus 4.8 (gloo-anthropic-claude-opus-4.8), Anthropic’s latest at the time, with fast variants for speed-sensitive work; Claude Opus 4.7 Fast and 4.6 Fast
  • GPT-5.3 Codex (gloo-openai-gpt-5.3-codex), a reasoning-focused coding model with a 400K context window
  • Qwen 3.5 Plus (gloo-qwen-3.5-plus-20260420), Qwen 3.5 27B (gloo-qwen-3.5-27b) and Qwen 3 235B A22B (gloo-qwen-3-235b-a22b-2507, gloo-qwen-3-235b-a22b-thinking-2507), including thinking variants
  • GLM-4.7 Flash (gloo-z-ai-glm-4.7-flash), built for fast inference
  • MiMo V2 Flash and MiMo V2.5 (gloo-xiaomi-mimo-v2.5), for reasoning, coding and agent tasks
  • Gemini 3.1 Flash Lite (gloo-google-gemini-3.1-flash-lite), now generally available
• 9 models added, week of May 18:
  • Gemini 3.5 Flash (gloo-google-gemini-3.5-flash, fast, strong at coding) and Qwen 3.7 Max (gloo-qwen-3.7-max, agent work) take a million tokens
  • Gemma 4 31B (gloo-google-gemma-4-31b), Google’s open-weight model with vision
  • GPT OSS 20B (gloo-openai-gpt-oss-20b), OpenAI’s open-weight model for consumer hardware
  • Ministral 3B, 8B and 14B (gloo-mistral-ministral-3b, gloo-mistral-ministral-8b, gloo-mistral-ministral-14b), spanning edge to server
  • Qwen 3.5 397B and 122B (gloo-qwen-3.5-397b-a17b, gloo-qwen-3.5-122b-a10b), large mixture-of-experts models
• 20+ models added, April 21 – May 15, including GPT-5.5 and GPT-5.5 Pro (gloo-openai-gpt-5.5, gloo-openai-gpt-5.5-pro), GPT-5.1 (gloo-openai-gpt-5.1), Claude Opus 4.1 (gloo-anthropic-claude-opus-4.1), DeepSeek V4 Flash and V4 Pro (gloo-deepseek-v4-flash, gloo-deepseek-v4-pro), DeepSeek R1 0528 (gloo-deepseek-r1-0528), Devstral 2 (gloo-mistral-devstral-2512), Qwen 3.6 Max Preview (gloo-qwen-3.6-max-preview) and Kimi K2.6 (gloo-kimi-k2.6). Kimi K2.6 takes images. Claude Opus 4.1, Devstral 2 and Qwen 3.6 Max Preview have since been redirected.

​

April 2026

• 11 models added, February 11 – April 20: Claude Opus 4.6 (gloo-anthropic-claude-opus-4.6), Claude Sonnet 4.6 (gloo-anthropic-claude-sonnet-4.6), Claude Opus 4.7 (gloo-anthropic-claude-opus-4.7), GPT-5.2 Pro (gloo-openai-gpt-5.2-pro), GPT-5.4 (gloo-openai-gpt-5.4), GPT-4.1 (gloo-openai-gpt-4.1), GPT-4.1 Mini (gloo-openai-gpt-4.1-mini), GPT-5 Nano (gloo-openai-gpt-5-nano), Meta Llama 4 Maverick (gloo-meta-llama-4-maverick), DeepSeek V3.2 (gloo-deepseek-v3.2) and DeepSeek R1 (gloo-deepseek-r1). Claude Opus 4.7 was available the same day Anthropic released it.

​

March 2026

• Gemini 3 Pro Preview (gloo-google-gemini-3-pro-preview): choose a current Gemini model from Supported Models.

​

February 2026

• 3 models added, January 25 – February 10: Claude Sonnet 4 (gloo-anthropic-claude-sonnet-4), Meta Llama 4 Maverick (gloo-meta-llama-4-maverick) and DeepSeek R1 (gloo-deepseek-r1). Claude Sonnet 4 and DeepSeek R1 have since been redirected.

​

January 2026

• 5 models: they no longer served requests and were removed from the catalog.

​

Next steps