Quick Facts
| Developer | Mistral AI |
| License | Modified MIT -- commercial license may be required above $20M consolidated global revenue; verify against the official model card |
| Typical Sizes | 123B -- cloud tag devstral-2:123b-cloud; local tag devstral-2:123b (~75 GB) also available on Ollama |
| Context | 256K tokens, native tool support |
| Modality | Text only -- no vision/image input |
| Best For | Agentic coding, deep codebase exploration, multi-file editing, software engineering agents |
| Recommended Deployment | Terminal.Glass Cloud tier (high usage) for evaluation; local GPU deployment for sustained heavy use |
Why Choose This Model
Devstral 2 is purpose-built for agentic software engineering -- using tools to explore a codebase, reading across many files, and making coordinated multi-file edits within a single agent loop. At 123B parameters it outperforms Mistral's smaller Devstral Small 2 on coding benchmarks, making it a strong choice for demanding refactors, cross-repository analysis, and complex debugging tasks. Cloud access lets you evaluate this capability immediately, without first sizing and provisioning the multi-GPU hardware a ~75 GB local checkpoint requires.
Terminal.Glass Deployment
Your interface, chat history, user accounts, and RAG index stay on your Terminal.Glass host.
Each request to devstral-2:123b-cloud
-- including code excerpts, repository context, and tool outputs -- is sent to Ollama Cloud for
inference and streamed back. This tag is billed at Ollama's high-usage tier, with cost scaling for
long 256K contexts and repeated agentic tool loops. No local GPU is required for the cloud tag; a
small VM or mini-PC running Open WebUI and the Ollama client is sufficient.
Recommended Uses
- Deep codebase exploration and cross-file reasoning on large repositories
- Multi-file edits and coordinated refactors driven by an agent loop
- Tool-assisted terminal and CI workflows for complex debugging
- Evaluating full-scale agentic coding before investing in local GPU hardware
Not recommended for proprietary source code, client repositories, credentials, or regulated codebases -- all request content leaves your network for this tag. Not suitable for vision or image-based tasks; this release is text-only. Generated multi-file edits and terminal-agent actions still require human review, tests, and rollback discipline before merging.
Hardware Guidance
For the cloud tag, the Terminal.Glass host only needs to run Open WebUI, the Ollama client, and your
document index -- a small VM or mini-PC is enough, since inference happens on Ollama Cloud. Devstral 2
is also available as a local weight release (devstral-2:123b,
~75 GB) for teams with sufficient GPU memory and storage to host it directly -- a practical option once
usage is steady and predictable. Confirm current tags at
the official Ollama listing.
Honest Guidance
- Regulated or confidential codebases: use the local
devstral-2:123btag on hardware you control instead of the cloud tag. - Steady daily team usage: local deployment on owned GPU hardware may cost less over time than recurring high-usage cloud tokens.
- Lighter coding workloads or tighter budgets: Devstral Small 2 (24B) offers a lower-cost cloud tier when full 123B capacity isn't needed.
- Vision or multimodal needs: this release is text-only; choose a different model family.
- Version pinning or audit reproducibility: cloud tags can change server-side; the local tag offers a pinned, reproducible alternative.
- Unreviewed agent automation: multi-file edits, generated patches, and terminal-agent actions need human review, tests, and rollback plans regardless of model choice.
Learn More
Full technical reference and licensing detail for the Devstral family are documented on NoCloudGPT. Confirm current tags and availability at ollama.com/library/devstral-2.