Quick start
ollama run devstral-small-2:latestRunnable variants
Memory is a rough file-size estimate; context length, runtime and GPU offload change it.
6 of 6 matching variants shown.6 of 6 variants shown.
| Tag | Type | Inputs | Digest | Download | Parameters | Quantization | Context | Est. memory | Run |
|---|---|---|---|---|---|---|---|---|---|
| 24b | Typelocal | Inputstext, image | Digest24277f07f62d | Download15 GB | Parameters24b | QuantizationUnknown | Context384K | Est. memory≈ 18 GB | ollama run devstral-small-2:24b |
| 24b-cloud | Typecloud | Inputstext, image | Digestec4a591da58a | DownloadUnknown | Parameters24b | QuantizationUnknown | Context256K | Est. memoryUnknown | ollama run devstral-small-2:24b-cloud |
| 24b-instruct-2512-fp16 | Typelocal | Inputstext, image | Digest15c77ff5438a | Download48 GB | Parameters24b | QuantizationFP16 | Context384K | Est. memory≈ 55 GB | ollama run devstral-small-2:24b-instruct-2512-fp16 |
| 24b-instruct-2512-q4_K_M | Typelocal | Inputstext, image | Digest24277f07f62d | Download15 GB | Parameters24b | QuantizationQ4_K_M | Context384K | Est. memory≈ 18 GB | ollama run devstral-small-2:24b-instruct-2512-q4_K_M |
| 24b-instruct-2512-q8_0 | Typelocal | Inputstext, image | Digest691a8e03a6dd | Download26 GB | Parameters24b | QuantizationQ8_0 | Context384K | Est. memory≈ 30.5 GB | ollama run devstral-small-2:24b-instruct-2512-q8_0 |
| latestdefault | Typelocal | Inputstext, image | Digest24277f07f62d | Download15 GB | ParametersUnknown | QuantizationUnknown | Context384K | Est. memory≈ 18 GB | ollama run devstral-small-2:latest |
Source-listed applications
Claude CodeHermes AgentOpenClawOpenCode
Related models
mistral-small3.2local
An update to Mistral Small that improves on function calling, instruction following, and less repetition errors.
mistral-small3.1local
Building upon Mistral Small 3, Mistral Small 3.1 (2503) adds state-of-the-art vision understanding and enhances long context capabilities up to 128k tokens without compromising text performance.
qwen3.6local
Qwen3.6 delivers substantial upgrades in agentic coding and thinking preservation than previous Qwen models.
muse-glimmerlocal
Meta's latest open model built for always-on local agents. 30B parameters, licensed under Apache 2.0 and runs on a single GPU — tuned for tool use, long tasks, and failure recovery.