Magistral Small 1.1
- In Input modalities: Text
- Reasoning
- Tool use
- Open weights
Prices
0 offerings · prices as listedNo priced offering is listed for this model yet.
Details
- Licence
- Apache 2.0
- Family
- magistral-small
Not stated on Mistral AI’s pages: max input, output modalities, reasoning efforts, reasoning can be off, max thinking budget, structured output, built-in tools, knowledge cutoff, aliases.
Run locally
Weights: mistralai/Magistral-Small-2507- Parameters 23.6B
- Dense
- Native precision BF16
- Layers 40
- KV cache per token 160 KB
- Config positions 40K (config.json max_position_embeddings)
- API context 39K (maker’s API)
- Needs
- ≈ 17.6 GB (weights 15.6 (maker’s file) + KV cache 1.3 + overhead 0.8)
- Usable
- 21.6 GB (90% of its memory)
- Generation speed
- 28–42 tok/s
- Longest context
- 33K (longer doesn’t fit)
- First token
- about 0.76–1.14 s for a 2K prompt
| Precision | Weights | Needs at 8K | On your hardware | Longest context |
|---|---|---|---|---|
| BF16 (native)maker’s file | 87.8 GB | ≈ 93.5 | ✕Doesn’t fit | — |
| Q8_0maker’s file | 23.3 GB | ≈ 25.8 | ○Needs system memory 7–10 tok/s | — |
| FP8 | 22.0 GB | ≈ 24.3 | ○Needs system memory 9–14 tok/s | — |
| Q6_K | 18.0 GB | ≈ 20.2 | ●Runs well 24–37 tok/s | 17K |
| Q5_K_Mmaker’s file | 15.6 GB | ≈ 17.6 | ●Runs well 28–42 tok/s | 33K |
| Q5_0 | 15.1 GB | ≈ 17.1 | ●Runs well 29–43 tok/s | 37K |
| Q4_K_Mmaker’s file | 13.3 GB | ≈ 15.3 | ●Runs well 32–48 tok/s | 39K |
| Q4_0 | 12.3 GB | ≈ 14.2 | ●Runs well 35–52 tok/s | 39K |
| IQ4_XS | 11.7 GB | ≈ 13.5 | ●Runs well 36–55 tok/s | 39K |
| Q3_K_Mquality drops | 10.7 GB | ≈ 12.5 | ●Runs well 39–59 tok/s | 39K |
| IQ3_XXSquality drops | 8.4 GB | ≈ 10.1 | ●Runs well 49–73 tok/s | 39K |
| Q2_Kquality drops | 8.2 GB | ≈ 10.0 | ●Runs well 49–74 tok/s | 39K |
| IQ2_XXSquality drops | 5.7 GB | ≈ 7.4 | ●Runs well 68–102 tok/s | 39K |
Click a row to choose a precision. “Maker’s file” sizes are the maker’s own files; others are parameters × bits per weight.
Compared with other hardware
Each at the highest precision it runs. Click one for details.
Smallest that runs it (Q4_K_M, 8K): GPU GeForce RTX 4090; unified memory Apple M5 Pro (16-core GPU) · 24 GB.
Run commands
llama-server -hf mistralai/Magistral-Small-2507-GGUF:Q5_K_M -c 8192
ollama run hf.co/mistralai/Magistral-Small-2507-GGUF:Q5_K_M
Memory to fine-tune
Batch 1, 2,048-token sequences, gradient checkpointing; LoRA and QLoRA train rank-16 adapters on the attention projections.
Estimates; facts from the maker’s Hugging Face repositories, as of 2026-11-06. How it’s estimated
Cited scores
From third-party leaderboards, not measured by usThis model isn’t on any leaderboard we cite yet.
Changes
All changes →No changes recorded yet.
Sources
- Models Overview | Mistral Docs (opens in a new tab) docs.mistral.ai · checked Nov 1, 2026
- Changelog | Mistral Docs (opens in a new tab) docs.mistral.ai · checked Nov 2, 2026
- Magistral Small 1.1 - Mistral AI | Mistral Docs (opens in a new tab) docs.mistral.ai · checked Nov 4, 2026
- Magistral Small 1.1 (opens in a new tab) huggingface.co · checked Nov 2, 2026
- Magistral Small 1.1 (opens in a new tab) docs.mistral.ai · checked Nov 5, 2026
- mistralai/Magistral-Small-2507 model card (opens in a new tab) huggingface.co · checked Nov 2, 2026
- mistralai/Magistral-Small-2507 config.json (opens in a new tab) huggingface.co · checked Nov 3, 2026
- mistralai/Magistral-Small-2507 files (opens in a new tab) huggingface.co · checked Nov 2, 2026
- mistralai/Magistral-Small-2507-GGUF files (opens in a new tab) huggingface.co · checked Nov 1, 2026
Mistral AI’s official policies: data use, privacy, terms →
Some values on this page are still waiting to be checked against official pages: a filled dot is verified, a hollow one is not yet (with or without a source on file). How we verify data
Mistral AI · France · mistral.ai
Related models
Snapshots, the same family and other models from this provider