Devstral 2 2512
Mistral AI · released Nov 28, 2025 · mistralai/Devstral-2-123B-Instruct-2512
- Type
- Open weightsCustom licence
- Params
- 125B
- Context
- 262K
about 197K words of context · download allowed, licence restricts use
Our take
Written Sep 30, 2026Devstral 2 2512 is a coding model you can download and run yourself, built for fixing issues inside an existing codebase. Its measured results point in two directions: strong on real GitHub issues, near the bottom of a board of human votes on web-app building.
Reach for it when the job is fixing issues in an existing project, where its SWE-bench Verified result is the relevant evidence, or when long files and large documents need to go in without being split up first. The licence puts conditions on commercial use and redistribution, so read it before you build on it. Skip it if you need a model that scores well on web-app building tasks judged by human preference, or if those licence conditions are a problem for your product.
The case for it
- 53.8% of real GitHub issues resolved end-to-end on SWE-bench Verified, placing 26th of 42 on that board as of 19 Feb 2026 — evidence for fixing issues in an existing project rather than set-piece exercises.
- Long documents and large code files fit in a single request without being split up first, though reliable recall across all of it is unverified in our data.
The case against it
- 93rd of 95 on Arena Code (WebDev) as of 25 Sep 2026, a board of human votes on web-app building tasks, so preference for its web-app output is among the weakest we list.
- The custom licence puts conditions on commercial use and redistribution, so it needs reading before you build on it.
- At 125 billion parameters this is not a download for a machine of your own; the fit verdict below is the check that settles it.
How good is it?
EverydayGeneral questions and everyday reasoning
Not yet scored on Arena Text (overall).
CodingWriting and fixing code on its own
Not yet scored on Arena Coding. It is on Arena Code (WebDev), in 93rd of 95 with 1196.
AgenticPlanning, calling tools, staying on task
Not yet scored on Arena Agent. It is on SWE-bench Verified, in 26th of 42 with 53.8.
WritingDrafting and rewriting prose
Not yet scored on Arena Creative Writing.
Every published score for this model2 scoresEvery figure we hold, from 2 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Can you run it yourself?
GeForce RTX 4090 · 24 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Apple M1 Pro (16-core GPU) · 32 GB
Too large for this card. The weights do not fit even with part of them offloaded to system memory.
Comfortable fit
Apple M1 Ultra (64-core GPU) · 128 GB
Room to spare. 12.9 GB spare means a 10% error in the size would not change the answer.
Memory use by level
Against a 24 GB card.
What is quantisation? →This model on every device we track71 devicesThe Q4 build most people download, on each device: what the weights come to, how much context the memory leaves, and whether it runs. Smallest device that runs it first.
Check against your own machine → · Where to rent it hosted →
Or rent it from someone else
Prices checked 2 hours ago — each listing carries its own date.
- per 1M tokens
- $0.40 in / $2.00 out
- Context served
- 262K
- Throughput
- ~61 tok/s
| Provider | In / out per 1M tokens | Context | Throughput | Trains on prompts | Logs prompts | Zero retention |
|---|---|---|---|---|---|---|
| OpenRouterOpenRouter's own listing | $0.40 / $2.00checked 2 hours ago | 262K | not measured | Unknown | Unknown | Unknown |
| Mistral AIeuThrough OpenRouter | $0.40 / $2.00checked 2 hours ago | 262K210K max reply | 61 tok/s | No | Yes30 days | Confirmed |
OpenRouterAcross the 2 listings we hold: 1 says it does not train on prompts, 0 say they do and 1 does not say. 1 appears in the zero-retention registry we check; the rest are unknown to us.
What each host's API supports
From the parameter list each endpoint publishes. Streaming is omitted: nothing we hold reports it, for any model.
| Provider | Tool calling | JSON output | Strict schema |
|---|---|---|---|
| OpenRouterOpenRouter's own listing | ✓ | ✓ | ✓ |
| Mistral AIeuThrough OpenRouter | ✓ | ✓ | ✓ |
Tool calling: 2 of 2 listings say yes. JSON output: 2 of 2 listings say yes. Strict schema: 2 of 2 listings say yes.
When we formed this view
Recent changes
What moved
input −9% ($0.44 → $0.40 per 1M tokens), output −9% ($2.20 → $2.00 per 1M tokens), cache read −9% ($0.044 → $0.040 per 1M tokens)What moved
first indexed by our pipelineEach date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- 1 of 2 listings does not say whether it trains on prompts.
- We hold no batch or off-peak rate for any of its listings.
- We hold a decode speed for it, but no prompt-processing (prefill) figure, so how long the input side of a job takes is unknown to us.
Licence and identifiers
What the licence allowsCustom licence, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.
Licence
Custom licence
This model ships custom license terms that don't map to a known template. We haven't parsed them, so commercial use, redistribution and derivatives are unverified — review the original terms before shipping.
Identifiers
- Hugging Face
- mistralai/Devstral-2-123B-Instruct-2512
- Architecture
- Dense
- Takes in, gives back
- Text and documents in, text out
- Catalogue slug
- mistralai-devstral-2-2512