Model Tracker
Mixtral 8x22B
Mixtral 8x22B is an AI model made by Mistral AI. Mistral AI released it on 17 April 2024. It is retired. Mistral AI retired it on 30 March 2025.
In inglese
Mixtral 8x22B is an AI model made by Mistral AI. Mistral AI released it on 17 April 2024. It is retired. Mistral AI retired it on 30 March 2025. It takes text and returns text. Its context window is 64,000 tokens. Its weights are open. Its licence is Apache 2.0.
- Released
- 17 April 2024 — mistral.ai, read 16 October 2026
- Status
- Retired — docs.mistral.ai, read 16 October 2026
- Retired
- 30 March 2025 — docs.mistral.ai, read 16 October 2026
- Takes
- Text — huggingface.co, read 16 October 2026
- Returns
- Text — huggingface.co, read 16 October 2026
- Open weights
- Yes — mistral.ai, read 16 October 2026
- Licence
- Apache 2.0 — mistral.ai, read 16 October 2026
- Context window
- 64,000 tokens — mistral.ai, read 16 October 2026
- Longest answer
- Not confirmed: The maker's model page and model card do not state the standard maximum output-token limit. The model-card generation example's max_new_tokens value is an example setting, not a published standard limit.
- Knowledge cutoff
- Not confirmed: The maker's model page, release announcement, and model card do not state a knowledge-cutoff date.
- API price, input
- Not confirmed: The maker's current API pricing page does not list Mixtral 8x22B, and the model is marked retired; no standard input price is confirmed there.
- API price, output
- Not confirmed: The maker's current API pricing page does not list Mixtral 8x22B, and the model is marked retired; no standard output price is confirmed there.
- Where
- Downloadable weights — docs.mistral.ai, read 16 October 2026
- Parameters
- Not confirmed: the figure is not printed in its sentence (a sum or a figure the page does not give)
- Model card
- Not confirmed: two readings disagree: https://huggingface.co/mistralai/Mixtral-8x22B-v0.1 and https://huggingface.co/mistralai/Mixtral-8x22B-Instruct-v0.1
Benchmarks
Measured by others, not by the maker: each score with the name the measurer gives the model and the setting it ran, its day and its licence.
Epoch AI
- Epoch Capabilities Index, Mixtral 8x22B: 122.1 (115.2–124.6), as of 19 October 2026
- GPQA Diamond, open-mixtral-8x22b: 34.1% (± 2.1), run 27 January 2025
Epoch AI, ‘Capabilities & benchmarking’, epoch.ai/benchmarks, CC BY 4.0: Epoch AI’s own runs and its Capabilities Index. Data · CC BY 4.0
Questions
When was Mixtral 8x22B released?
Mistral AI released Mixtral 8x22B on 17 April 2024.
Is Mixtral 8x22B still available?
No: Mistral AI retired Mixtral 8x22B on 30 March 2025.
What is the context window of Mixtral 8x22B?
The context window of Mixtral 8x22B is 64,000 tokens, as Mistral AI prints it.
Are the weights of Mixtral 8x22B open?
Yes: Mistral AI publishes the weights of Mixtral 8x22B for download.
What is the licence of Mixtral 8x22B?
The licence of Mixtral 8x22B is Apache 2.0, as Mistral AI names it.
What does Mixtral 8x22B take and return?
Mixtral 8x22B takes text and returns text.
More from Mistral AI
- Codestral
- Codestral Mamba 7B
- Devstral 2
- Devstral Medium 1.0
- Devstral Small 1.0
- Devstral Small 1.1
- Devstral Small 2
- Leanstral
- Leanstral 1.5
- Magistral Medium 1.0
- Magistral Medium 1.1
- Magistral Medium 1.2
- Magistral Small 1.0
- Magistral Small 1.1
- Magistral Small 1.2
- Mathstral 7B
- Ministral 3 14B
- Ministral 3 8B
- Ministral 8B
- Mistral 7B
- Mistral Large 1.0
- Mistral Large 2.0
- Mistral Large 2.1
- Mistral Large 3
Sources
- Mixtral 8x22B - Mistral AI | Mistral Docs (read 16 October 2026)
- Cheaper, Better, Faster, Stronger (read 16 October 2026)
- mistralai/Mixtral-8x22B-v0.1 (read 16 October 2026)
- API pricing (read 16 October 2026)