ThinkFacility

Model Watch: every model we track, and what it costs

One row per model from Alibaba, Anthropic, DeepSeek, Google, Meta, Mistral, OpenAI and xAI. What it is, when it shipped, how big its context is and what a million tokens costs, taken from the labs' own documentation. Then every time the tracker has seen the name.

· 61 models · release timeline

ModelLabFromStatusDateContextPrice per 1M tokensSignals
GPT-6 AstraOpenAIUSreleased3 September 20261050K tokens$10.00 in, $50.00 out16
Muse Spark 1.3MetaUSreleased2 September 20261,048,576 tokens$1.25 in, $4.25 out2
Gemini 3.8 FlashGoogleUSreleased2 September 20261,048,576 tokens$0.75 in, $3.75 out2
Claude Mythos 5.1AnthropicUSreleased1 September 20261M tokens$10.00 in, $50.00 out4
Claude Fable 5.1AnthropicUSreleased1 September 20261M tokens$10.00 in, $50.00 out41
DeepSeek-V4-Flash-Vision-ExpDeepSeekChinapreview21 August 20261M tokens$0.44 in, $1.32 out1
Gemini 3.7 FlashGoogleUSreleased13 August 20261,048,576 tokens$0.75 in, $3.75 out3
DeepSeek-V4-ProDeepSeekChinareleased13 August 20261M tokens$1.32 in, $3.96 out4
Grok 4.6xAIUSreleased12 August 2026500K tokens$2.00 in, $6.00 out7
Muse GlimmerMetaUSreleased10 August 2026131,072 tokens—1
Qwen3.8-27BAlibabaChinareleased5 August 2026——3
DeepSeek-V4-FlashDeepSeekChinareleased31 July 20261M tokens$0.44 in, $1.32 out5
Claude Opus 5AnthropicUSreleased24 July 20261M tokens$5.00 in, $25.00 out18
Gemini 3.6 FlashGoogleUSreleased21 July 20261,048,576 tokens$0.75 in, $3.75 out3
Gemini 3.5 Flash-LiteGoogleUSreleased21 July 20261,048,576 tokens$0.30 in, $2.50 out0
Grok 4.5xAIUSreleased16 July 2026500K tokens$2.00 in, $6.00 out4
Muse Spark 1.1MetaUSreleased9 July 20261,048,576 tokens$1.25 in, $4.25 out1
GPT-5.6 TerraOpenAIUSreleased9 July 20261050K tokens$2.00 in, $12.00 out0
GPT-5.6 SolOpenAIUSreleased9 July 20261050K tokens$4.00 in, $20.00 out5
GPT-5.6 LunaOpenAIUSreleased9 July 20261050K tokens$0.20 in, $1.20 out2
Leanstral 1.5MistralFrancepreview30 June 2026256K tokensFree2
Claude Sonnet 5AnthropicUSreleased30 June 20261M tokens$2.00 in, $10.00 out10
Claude Fable 5AnthropicUSreleased9 June 20261M tokens$10.00 in, $50.00 out9
Grok Build 0.1xAIUSreleased29 May 2026256K tokens$1.00 in, $2.00 out1
Gemini 3.5 FlashGoogleUSreleased19 May 20261,048,576 tokens$1.50 in, $9.00 out2
Gemini 3.1 Flash-LiteGoogleUSreleased7 May 20261,048,576 tokens$0.25 in, $1.50 out0
Mistral Medium 3.5MistralFrancereleased28 April 2026256K tokens$1.50 in, $7.50 out2
GPT-5.5OpenAIUSreleased23 April 20261050K tokens$5.00 in, $30.00 out0
GPT-5.4 nanoOpenAIUSreleased17 March 2026400K tokens$0.20 in, $1.25 out0
GPT-5.4 MiniOpenAIUSreleased17 March 2026400K tokens$0.75 in, $4.50 out0
Mistral Small 4MistralFrancereleased16 March 2026256K tokens$0.15 in, $0.60 out4
GPT-5.4OpenAIUSreleased5 March 20261050K tokens$2.50 in, $15.00 out0
Gemini 3.1 Pro PreviewGoogleUSpreview19 February 20261,048,576 tokens$2.00 in, $12.00 out0
GPT-5.3-CodexOpenAIUSreleased5 February 2026400K tokens$1.75 in, $14.00 out0
Gemini 3 Flash PreviewGoogleUSpreview17 December 20251,048,576 tokens$0.50 in, $3.00 out0
GPT-5.2OpenAIUSreleased11 December 2025400K tokens$1.75 in, $14.00 out0
Mistral Large 3MistralFrancereleased2 December 2025256K tokens$0.50 in, $1.50 out3
Ministral 3 14BMistralFrancereleased2 December 2025256K tokens$0.20 in, $0.20 out0
Claude Opus 4.5AnthropicUSreleased24 November 2025200K tokens$5.00 in, $25.00 out0
GPT-5.1OpenAIUSreleased13 November 2025400K tokens$1.25 in, $10.00 out1
Claude Haiku 4.5AnthropicUSreleased15 October 2025200K tokens$1.00 in, $5.00 out7
Claude Sonnet 4.5AnthropicUSreleased29 September 2025200K tokens$3.00 in, $15.00 out0
GPT-5OpenAIUSdeprecated7 August 2025400K tokens$1.25 in, $10.00 out1
CodestralMistralFrancereleased30 July 2025128K tokens$0.30 in, $0.90 out2
o3OpenAIUSdeprecated16 April 2025200K tokens$2.00 in, $8.00 out1
Qwen4AlibabaChinaannounced———0
Qwen3.8-MaxAlibabaChinareleased—1M tokens$2.00 in, $6.00 out0
Qwen3.8-FlashAlibabaChinareleased—1M tokens$0.15 in, $0.47 out3
Qwen3.7-PlusAlibabaChinareleased—1M tokens$0.40 in, $1.60 out0
Muse VideoMetaUSannounced———1
Muse Spark 1.2MetaUSreleased—1,048,576 tokens$1.25 in, $4.25 out0
Llama 5MetaUSrumoured———0
Llama 4 ScoutMetaUSreleased———1
Grok 5xAIUSrumoured———0
Grok 4.3xAIUSreleased—1M tokens$1.25 in, $2.50 out0
Gemini 3.5 ProGoogleUSannounced———0
Gemini 3.1 Deep ThinkGoogleUSreleased———0
Gemini 2.5 ProGoogleUSreleased—1,048,576 tokens$1.25 in, $10.00 out0
GPT-5.5 ProOpenAIUSreleased—1050K tokens$30.00 in, $180.00 out0
Claude Opus 5.1AnthropicUSrumoured———5
Claude Opus 4.1AnthropicUSdeprecated——$15.00 in, $75.00 out2

How to read this

Status we set by hand. Released means you can call it. Preview means you can call it and the lab reserves the right to change it. Announced means the lab has said it's coming. Rumoured means only other people are saying so.

Signals counts tracked posts and Reddit threads that name the model. It's how we decide which models are worth a page, and it's the first column that moves when something is about to happen.

Prices are list prices per million tokens from each lab's own docs. They skip caching and batch discounts, and a dash means the lab hasn't published a number. We'd rather leave it blank than guess.

Which models get their own page?

18 of them. A model earns a page when we've sourced at least 2 hard facts about it and the tracker has logged at least 3 mentions or threads. Below that there's nothing to say that this table doesn't already say, so we don't write a page nobody would want to read.

The other 43 sit here instead:

Qwen3.7-Plus

Alibaba · China · 1M tokens · $0.40 in, $1.60 out per 1M. The mid tier. Prompts over 256K tokens triple the input rate, so the sticker price only holds for short calls. Source.

0 signals, 4 of 5 specs sourced. A page needs 3 signals and 2 specs.

Qwen3.8-Max

Alibaba · China · 1M tokens · $2.00 in, $6.00 out per 1M. Alibaba's flagship, at a third of what Fable 5.1 costs. Prices are region-scoped: this is the Singapore column, and Beijing is cheaper again. Source.

0 signals, 4 of 5 specs sourced. A page needs 3 signals and 2 specs.

Qwen4

Alibaba · China. Alibaba has said the words out loud, in a Hugging Face README describing Qwen3.8-Flash-Next as a preview of the architecture behind Qwen4. That's the entire public record. Source.

0 signals, 1 of 5 specs sourced. A page needs 3 signals and 2 specs.

Claude Opus 4.5

Anthropic · US · released 24 November 2025 · 200K tokens · $5.00 in, $25.00 out per 1M. The last Opus before the million-token window. Same money as Opus 5 for a fifth of the context. Source.

0 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Claude Sonnet 4.5

Anthropic · US · released 29 September 2025 · 200K tokens · $3.00 in, $15.00 out per 1M. Costs more than Sonnet 5 and gives you less. It's here because a lot of production code still pins to it. Source.

0 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Claude Opus 5.1

Anthropic · US. There is no Opus 5.1. Opus 5 shipped in July and the 5.1 generation went out as Fable and Mythos, so this entry exists to catch the name if it ever turns up.

5 Reddit threads name it, the newest from 2 September 2026. They'll appear here in full once it clears the bar.

5 signals, 0 of 5 specs sourced. A page needs 3 signals and 2 specs.

Claude Opus 4.1

Anthropic · US · $15.00 in, $75.00 out per 1M. Retired on 5 August 2026. It survives on Bedrock and Google Cloud, and Anthropic dropped its context window off the docs when it went. Source.

2 signals, 3 of 5 specs sourced. A page needs 3 signals and 2 specs.

DeepSeek-V4-Flash-Vision-Exp

DeepSeek · China · released 21 August 2026 · 1M tokens · $0.44 in, $1.32 out per 1M. The multimodal one, and DeepSeek calls it experimental in its own release note. Priced like Flash. Source.

1 signal, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Gemini 2.5 Pro

Google · US · 1,048,576 tokens · $1.25 in, $10.00 out per 1M. Last generation, still stable, and cheaper than the 3.1 Pro preview. Google's page gives a month rather than a release date. Source.

0 signals, 4 of 5 specs sourced. A page needs 3 signals and 2 specs.

Gemini 3.1 Deep Think

Google · US. It exists, it's in the Gemini app, and that's all Google will tell you. No API id and no price anywhere on their docs. Source.

0 signals, 1 of 5 specs sourced. A page needs 3 signals and 2 specs.

Gemini 3.1 Flash-Lite

Google · US · released 7 May 2026 · 1,048,576 tokens · $0.25 in, $1.50 out per 1M. Cheaper than the 3.5 Lite that replaced it, and audio costs double the text rate. Worth knowing if your volume is text. Source.

0 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Gemini 3.5 Flash

Google · US · released 19 May 2026 · 1,048,576 tokens · $1.50 in, $9.00 out per 1M. Older and dearer than every Flash that followed it, because it never got the introductory rate. Move off it. Source.

2 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Gemini 3.5 Flash-Lite

Google · US · released 21 July 2026 · 1,048,576 tokens · $0.30 in, $2.50 out per 1M. Thirty cents a million on a million-token window. The cheapest Gemini you can call that isn't a generation behind. Source.

0 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Gemini 3.8 Flash

Google · US · released 2 September 2026 · 1,048,576 tokens · $0.75 in, $3.75 out per 1M. Google's newest stable model, and it's a Flash. The price is introductory: it doubles on 1 January 2027, so budget for $1.50 in and $7.50 out. Source.

1 Reddit thread name it, the newest from 3 September 2026. They'll appear here in full once it clears the bar.

2 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Gemini 3 Flash Preview

Google · US · released 17 December 2025 · 1,048,576 tokens · $0.50 in, $3.00 out per 1M. The first Gemini 3 Flash, still sitting in preview while four stable ones shipped past it. Source.

0 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Gemini 3.1 Pro Preview

Google · US · released 19 February 2026 · 1,048,576 tokens · $2.00 in, $12.00 out per 1M. The only Pro-tier Gemini you can call, and Google still labels it preview seven months in. Prompts over 200K tokens bill at double. Source.

0 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Gemini 3.5 Pro

Google · US. Google says "3.5 Pro coming soon" and nothing else. No id, no price, no date, and the Flash line has already run four versions past it. Source.

0 signals, 1 of 5 specs sourced. A page needs 3 signals and 2 specs.

Llama 4 Scout

Meta · US. The newest Llama, and it's from April 2025. Meta's docs quote a 10M context window without giving a token count, and there's no first-party price because the weights are the product. Source.

1 signal, 1 of 5 specs sourced. A page needs 3 signals and 2 specs.

Muse Glimmer

Meta · US · released 10 August 2026 · 131,072 tokens. Meta's open-weight agentic model, Apache 2.0, and the only Muse you can run yourself. Note the org: it's meta-models on Hugging Face, not meta-llama. Source.

1 Reddit thread name it, the newest from 26 August 2026. They'll appear here in full once it clears the bar.

1 signal, 3 of 5 specs sourced. A page needs 3 signals and 2 specs.

Muse Spark 1.1

Meta · US · released 9 July 2026 · 1,048,576 tokens · $1.25 in, $4.25 out per 1M. The first Muse Spark and the launch of Meta's own model API. It went out as a public preview and is priced the same as the two that followed. Source.

1 signal, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Muse Spark 1.2

Meta · US · 1,048,576 tokens · $1.25 in, $4.25 out per 1M. Same price and window as 1.3. Meta never posted a launch date for it, only an evaluation follow-up in August, so we've left the date blank. Source.

0 signals, 4 of 5 specs sourced. A page needs 3 signals and 2 specs.

Muse Spark 1.3

Meta · US · released 2 September 2026 · 1,048,576 tokens · $1.25 in, $4.25 out per 1M. Meta's frontier model ships under Muse now, not Llama. A contributor tier drops the input price to ten cents a million if you qualify. Source.

2 Reddit threads name it, the newest from 2 September 2026. They'll appear here in full once it clears the bar.

2 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Muse Video

Meta · US. Announced alongside Muse Image in July and still described as coming soon. No API, no price. Source.

1 signal, 1 of 5 specs sourced. A page needs 3 signals and 2 specs.

Llama 5

Meta · US. There is no Llama 5. The meta-llama org on Hugging Face last published in November 2025, and Meta's 2026 open-weight release went out under a different org entirely.

0 signals, 0 of 5 specs sourced. A page needs 3 signals and 2 specs.

Codestral

Mistral · France · released 30 July 2025 · 128K tokens · $0.30 in, $0.90 out per 1M. The code model. It's a year old now and Mistral hasn't refreshed it since folding Devstral away. Source.

2 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Ministral 3 14B

Mistral · France · released 2 December 2025 · 256K tokens · $0.20 in, $0.20 out per 1M. Input and output cost the same, which is unusual and makes it easy to budget for chatty workloads. Source.

0 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Mistral Medium 3.5

Mistral · France · released 28 April 2026 · 256K tokens · $1.50 in, $7.50 out per 1M. Mistral's frontier model, and the weights are out under a modified MIT licence. The whole Magistral and Devstral lines were folded into this one. Source.

2 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Leanstral 1.5

Mistral · France · released 30 June 2026 · 256K tokens · Free per 1M. Free while it's in preview, Apache 2.0, 256K in and 128K out. The catch is that a preview price is a preview price. Source.

2 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

GPT-5.1

OpenAI · US · released 13 November 2025 · 400K tokens · $1.25 in, $10.00 out per 1M. Cheaper than everything newer except the small models, and it's still the model OpenAI's own docs describe as best for coding. Read that as stale copy, not a recommendation. Source.

1 signal, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

GPT-5.2

OpenAI · US · released 11 December 2025 · 400K tokens · $1.75 in, $14.00 out per 1M. December's flagship, still callable. Its chat variant was shut down in August; the base model wasn't. Source.

0 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

GPT-5.3-Codex

OpenAI · US · released 5 February 2026 · 400K tokens · $1.75 in, $14.00 out per 1M. The last Codex model still standing. Every earlier one in that line was shut down on 23 July 2026. Source.

0 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

GPT-5.4

OpenAI · US · released 5 March 2026 · 1050K tokens · $2.50 in, $15.00 out per 1M. March's flagship, still up and still cheaper than GPT-5.5. OpenAI's own catalogue blurbs haven't caught up with the 5.6 line. Source.

0 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

GPT-5.4 Mini

OpenAI · US · released 17 March 2026 · 400K tokens · $0.75 in, $4.50 out per 1M. The mid-size option on a 400K window. Luna undercuts it heavily, so read both prices before you pick. Source.

0 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

GPT-5.4 nano

OpenAI · US · released 17 March 2026 · 400K tokens · $0.20 in, $1.25 out per 1M. Same input price as Luna with a smaller window and a smaller model behind it. Source.

0 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

GPT-5.5

OpenAI · US · released 23 April 2026 · 1050K tokens · $5.00 in, $30.00 out per 1M. April's flagship. It costs more than Sol and does less, which is what happens when a lab ships twice in a quarter. Source.

0 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

GPT-5.5 Pro

OpenAI · US · 1050K tokens · $30.00 in, $180.00 out per 1M. Six times the price of GPT-5.5 for the reasoning-heavy end. Check it actually beats Sol on your evals before you commit a budget to it. Source.

0 signals, 4 of 5 specs sourced. A page needs 3 signals and 2 specs.

GPT-5.6 Luna

OpenAI · US · released 9 July 2026 · 1050K tokens · $0.20 in, $1.20 out per 1M. Twenty cents a million in. This is the one to point your bulk classification and extraction jobs at. Source.

2 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

GPT-5.6 Terra

OpenAI · US · released 9 July 2026 · 1050K tokens · $2.00 in, $12.00 out per 1M. Half the input price of Sol on the same million-token window. Worth a try before you reach for the flagship. Source.

0 signals, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

GPT-5

OpenAI · US · released 7 August 2025 · 400K tokens · $1.25 in, $10.00 out per 1M. Shuts down on 11 December 2026, along with GPT-5 Mini and nano. If you pinned to it a year ago, that's your deadline. Source.

1 signal, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

o3

OpenAI · US · released 16 April 2025 · 200K tokens · $2.00 in, $8.00 out per 1M. The o-series is being wound down. o3 and o3-pro go on 11 December 2026, o3-mini and o4-mini on 23 October. Source.

1 signal, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Grok 4.3

xAI · US · 1M tokens · $1.25 in, $2.50 out per 1M. Twice the context of the flagship at a bit over half the price. xAI never posted a launch note for it, so there's no date to give you. Source.

0 signals, 4 of 5 specs sourced. A page needs 3 signals and 2 specs.

Grok Build 0.1

xAI · US · released 29 May 2026 · 256K tokens · $1.00 in, $2.00 out per 1M. The coding model, in public beta. The old grok-code-fast names are aliases onto this now. Source.

1 signal, 5 of 5 specs sourced. A page needs 3 signals and 2 specs.

Grok 5

xAI · US. No Grok 5. xAI is still numbering inside 4.x, and the lab is a SpaceX subsidiary now, which may explain why the roadmap went quiet.

0 signals, 0 of 5 specs sourced. A page needs 3 signals and 2 specs.

Keep reading

The release timeline is the same catalogue sorted by date, with filters for lab, country and status. The tracker is the raw feed everything here is built from.