Devstral 2 2512 vs GPT-4.1
GPT-4.1 leads Devstral 2 2512 on the Artificial Analysis Intelligence Index, 12.7 to 8.6, and Devstral 2 2512 costs 4.0x less per output token ($2.00 vs $8.00 per million). Updated October 21, 2026; 3 benchmarks have a result for both.
Devstral 2 2512 or GPT-4.1: which to pick
Neither model has a coding score yet, so the two cannot be ranked on coding.
Pick Devstral 2 2512 for output tokens 4.0x cheaper ($2.00 vs $8.00 per million) and work measured by LiveCodeBench (Vals AI) or AA Intelligence Index, where it leads.
- Shared results
- 3
- Devstral 2 2512 only
- 3
- GPT-4.1 only
- 89
- Like-for-like categories
- 2 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
No call
Neither has a coding score yet.
One side not measured
Intelligence
AA Intelligence Index
GPT-4.1
GPT-4.1 leads, 12.7 against 8.6.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only Devstral 2 2512 has a AA Agentic Index result, 2.3.
Directional only
Reasoning
GPQA Diamond
No call
Only GPT-4.1 has a GPQA Diamond result, 65.4%.
Directional only
Math
AIME
No call
Only GPT-4.1 has a AIME result, 39.6%.
Directional only
Knowledge
MMLU Pro
No call
Only GPT-4.1 has a MMLU Pro result, 80.5%.
Directional only
Web design
Design Arena: websites
No call
Only GPT-4.1 has a Design Arena: websites result, 1039.
Directional only
Long documents
Listed context window
GPT-4.1
GPT-4.1 takes 1.05M tokens against 262K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Devstral 2 2512
Devstral 2 2512 $0.0014, GPT-4.1 $0.0060 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
GPT-4.1
Devstral 2 2512 $1.19, GPT-4.1 $0.37.
Published attempt cost and pass rate
At a glance
- Coding index
- Devstral 2 2512 not benchmarked, GPT-4.1 not benchmarked.
- Widest category
- Coding, 8.5 points apart on average over 2 shared benchmarks.
- Widest single gap
- SWE-bench Verified: Devstral 2 2512 53.8% against GPT-4.1 39.6%, 14.2 points apart on a 0 to 100 scale.
- Output price
- Devstral 2 2512 $2.00 against GPT-4.1 $8.00 per million output tokens, 4.0x apart.
- Context window
- Devstral 2 2512 takes 262K tokens, GPT-4.1 1.05M.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | Devstral 2 2512 | Bars | GPT-4.1 | Ties |
|---|---|---|---|---|
| Coding | 1 | 1 | ||
| Intelligence | 0 | 1 | ||
| All shared boards | 1 | 2 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index11
| Benchmark | Devstral 2 2512 | GPT-4.1 | Leader |
|---|---|---|---|
| Aider polyglot | Not measured | 52.4% | Only GPT-4.1 measured |
| SWE-bench Verified | 53.8% | 39.6% | Devstral 2 2512 |
| SWE-bench Multimodal | Not measured | 31.1% | Only GPT-4.1 measured |
| AA Coding Index | 31.3 | Not measured | Only Devstral 2 2512 measured |
| LiveCodeBench (Vals AI) | 51.8% | 54.7% | GPT-4.1 |
| SWE-bench Verified (Vals AI) | 62.8% | Not measured | Only Devstral 2 2512 measured |
| LiveCodeBench (Artificial Analysis run) | Not measured | 45.7% | Only GPT-4.1 measured |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 13.6% | Only GPT-4.1 measured |
| WeirdML | Not measured | 39.0% | Only GPT-4.1 measured |
| ALE-Bench | Not measured | 558.1 | Only GPT-4.1 measured |
| LMArena Text: coding | Not measured | 1390 | Only GPT-4.1 measured |
| LMArena Text: software and IT | Not measured | 1391 | Only GPT-4.1 measured |
| SciCode | Not measured | 4.6% | Only GPT-4.1 measured |
| Neither model has a result on the other 26 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Devstral 2 2512 | GPT-4.1 | Leader |
|---|---|---|---|
| AA Intelligence Index | 8.6 | 12.7 | GPT-4.1 |
| Epoch Capabilities Index | Not measured | 136.7 | Only GPT-4.1 measured |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index4 boards, none shared
| Benchmark | Devstral 2 2512 | GPT-4.1 | Leader |
|---|---|---|---|
| AA Agentic Index | 2.3 | Not measured | Only Devstral 2 2512 measured |
| τ²-Bench Telecom | Not measured | 47.1% | Only GPT-4.1 measured |
| Berkeley Function Calling Leaderboard (BFCL) | Not measured | 54.0% | Only GPT-4.1 measured |
| BrowseComp-Plus | Not measured | 36.4% | Only GPT-4.1 measured |
| Neither model has a result on the other 37 boards here | |||
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index15 boards, none shared
| Benchmark | Devstral 2 2512 | GPT-4.1 | Leader |
|---|---|---|---|
| GPQA Diamond | Not measured | 65.4% | Only GPT-4.1 measured |
| MMMU Pro (Vals AI) | Not measured | 72.4% | Only GPT-4.1 measured |
| GPQA Diamond (Artificial Analysis run) | Not measured | 66.6% | Only GPT-4.1 measured |
| Humanity's Last Exam (Artificial Analysis run) | Not measured | 4.2% | Only GPT-4.1 measured |
| GPQA Diamond (Epoch AI run) | Not measured | 66.9% | Only GPT-4.1 measured |
| ARC-AGI | Not measured | 5.5% | Only GPT-4.1 measured |
| ARC-AGI-2 | Not measured | 0.4% | Only GPT-4.1 measured |
| SimpleBench | Not measured | 27.0% | Only GPT-4.1 measured |
| Chess puzzles | Not measured | 3.0% | Only GPT-4.1 measured |
| LMArena Text: hard prompts | Not measured | 1384 | Only GPT-4.1 measured |
| LMArena Text: expert | Not measured | 1364 | Only GPT-4.1 measured |
| LMArena Text: hard prompts in English | Not measured | 1393 | Only GPT-4.1 measured |
| LMArena Vision: diagrams | Not measured | 1215 | Only GPT-4.1 measured |
| LMArena Vision: homework | Not measured | 1248 | Only GPT-4.1 measured |
| Kagi LLM Benchmark | Not measured | 52.3% | Only GPT-4.1 measured |
| Neither model has a result on the other 7 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index10 boards, none shared
| Benchmark | Devstral 2 2512 | GPT-4.1 | Leader |
|---|---|---|---|
| AIME | Not measured | 39.6% | Only GPT-4.1 measured |
| MATH 500 (Vals AI) | Not measured | 87.2% | Only GPT-4.1 measured |
| MGSM (Vals AI) | Not measured | 87.7% | Only GPT-4.1 measured |
| AA Math Index | Not measured | 34.7 | Only GPT-4.1 measured |
| AIME 2025 (Artificial Analysis run) | Not measured | 34.7% | Only GPT-4.1 measured |
| OTIS Mock AIME | Not measured | 38.3% | Only GPT-4.1 measured |
| FrontierMath, tiers 1 to 3 | Not measured | 6.0% | Only GPT-4.1 measured |
| MATH level 5 | Not measured | 83.0% | Only GPT-4.1 measured |
| LMArena Text: math | Not measured | 1369 | Only GPT-4.1 measured |
| LMArena Text: mathematical work | Not measured | 1360 | Only GPT-4.1 measured |
| Neither model has a result on the other 3 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index15 boards, none shared
| Benchmark | Devstral 2 2512 | GPT-4.1 | Leader |
|---|---|---|---|
| MMLU Pro | Not measured | 80.5% | Only GPT-4.1 measured |
| LegalBench (Vals AI) | Not measured | 83.1% | Only GPT-4.1 measured |
| CaseLaw (Vals AI) | Not measured | 69.9% | Only GPT-4.1 measured |
| CorpFin (Vals AI) | Not measured | 63.0% | Only GPT-4.1 measured |
| TaxEval (Vals AI) | Not measured | 75.1% | Only GPT-4.1 measured |
| MortgageTax (Vals AI) | Not measured | 65.9% | Only GPT-4.1 measured |
| MedQA (Vals AI) | Not measured | 91.2% | Only GPT-4.1 measured |
| MMLU Pro (Artificial Analysis run) | Not measured | 80.6% | Only GPT-4.1 measured |
| SimpleQA Verified | Not measured | 31.1% | Only GPT-4.1 measured |
| GeoBench | Not measured | 72.0% | Only GPT-4.1 measured |
| LMArena Text: science | Not measured | 1382 | Only GPT-4.1 measured |
| LMArena Text: medicine and healthcare | Not measured | 1379 | Only GPT-4.1 measured |
| LMArena Text: legal and government | Not measured | 1399 | Only GPT-4.1 measured |
| LMArena Text: business and finance | Not measured | 1369 | Only GPT-4.1 measured |
| FACTS Benchmark Suite | Not measured | 45.8% | Only GPT-4.1 measured |
| Neither model has a result on the other 3 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index6 boards, none shared
| Benchmark | Devstral 2 2512 | GPT-4.1 | Leader |
|---|---|---|---|
| Design Arena: websites | Not measured | 1039 | Only GPT-4.1 measured |
| Design Arena: UI components | Not measured | 1000 | Only GPT-4.1 measured |
| Design Arena: game dev | Not measured | 1080 | Only GPT-4.1 measured |
| Design Arena: data viz | Not measured | 1104 | Only GPT-4.1 measured |
| Design Arena: 3D | Not measured | 850 | Only GPT-4.1 measured |
| Design Arena: all code categories | Not measured | 1026 | Only GPT-4.1 measured |
| Neither model has a result on the other 3 boards here | |||
GeneralDirectional onlyPublic boards, shown as published, kept out of the index30 boards, none shared
| Benchmark | Devstral 2 2512 | GPT-4.1 | Leader |
|---|---|---|---|
| IFBench | Not measured | 43.0% | Only GPT-4.1 measured |
| Long context reasoning (AA-LCR) | Not measured | 68.3% | Only GPT-4.1 measured |
| DTBench | Not measured | 68.3% | Only GPT-4.1 measured |
| LMCA | Not measured | 25.6% | Only GPT-4.1 measured |
| Fiction.LiveBench | Not measured | 63.9% | Only GPT-4.1 measured |
| LMArena Text | Not measured | 1383 | Only GPT-4.1 measured |
| LMArena Text: instruction following | Not measured | 1367 | Only GPT-4.1 measured |
| LMArena Text: creative writing | Not measured | 1363 | Only GPT-4.1 measured |
| LMArena Vision | Not measured | 1210 | Only GPT-4.1 measured |
| LMArena Text: English | Not measured | 1392 | Only GPT-4.1 measured |
| LMArena Text: non-English | Not measured | 1370 | Only GPT-4.1 measured |
| LMArena Text: Chinese | Not measured | 1382 | Only GPT-4.1 measured |
| LMArena Text: German | Not measured | 1381 | Only GPT-4.1 measured |
| LMArena Text: Japanese | Not measured | 1319 | Only GPT-4.1 measured |
| LMArena Text: Korean | Not measured | 1338 | Only GPT-4.1 measured |
| LMArena Text: French | Not measured | 1382 | Only GPT-4.1 measured |
| LMArena Text: Spanish | Not measured | 1376 | Only GPT-4.1 measured |
| LMArena Text: Russian | Not measured | 1377 | Only GPT-4.1 measured |
| LMArena Text: Polish | Not measured | 1389 | Only GPT-4.1 measured |
| LMArena Text: multi-turn | Not measured | 1397 | Only GPT-4.1 measured |
| LMArena Text: longer queries | Not measured | 1385 | Only GPT-4.1 measured |
| LMArena Text: writing and language | Not measured | 1371 | Only GPT-4.1 measured |
| LMArena Text: entertainment and media | Not measured | 1365 | Only GPT-4.1 measured |
| LMArena Vision: English | Not measured | 1209 | Only GPT-4.1 measured |
| LMArena Vision: Chinese | Not measured | 1226 | Only GPT-4.1 measured |
| LMArena Vision: reading text | Not measured | 1218 | Only GPT-4.1 measured |
| LMArena Vision: humor | Not measured | 1196 | Only GPT-4.1 measured |
| LMArena Vision: captioning | Not measured | 1207 | Only GPT-4.1 measured |
| LMArena Vision: recognizing things | Not measured | 1208 | Only GPT-4.1 measured |
| LMArena Vision: creative writing | Not measured | 1210 | Only GPT-4.1 measured |
| Neither model has a result on the other 16 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Devstral 2 2512 | GPT-4.1 | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Devstral 2 2512 costs less
Repository review
50K input + 3K output tokens
Devstral 2 2512 costs less
Agent loop
200K input + 10K output tokens
Devstral 2 2512 costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | Mistral Devstral 2 2512 | OpenAI GPT-4.1 |
|---|---|---|
| Maker | Mistral AI | OpenAI |
| Weights | Open weights | No public weights listed |
| Released | December 9, 2025 | April 14, 2025 |
| Context window | 262K | 1.05M |
| Input $/Mtok | $0.40 | $2.00 |
| Output $/Mtok | $2.00 | $8.00 |
| Coding index | Not benchmarked | Not benchmarked |
| Independent measurements behind it | 1 | 0 |
| Cost per solved task | $1.19 | $0.37 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Devstral 2 2512 better than GPT-4.1 for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. Pick Devstral 2 2512 for output tokens 4.0x cheaper ($2.00 vs $8.00 per million) and work measured by LiveCodeBench (Vals AI) or AA Intelligence Index, where it leads.
Which is cheaper, Devstral 2 2512 or GPT-4.1?
Devstral 2 2512 costs $0.40 per million input tokens and $2.00 per million output tokens; GPT-4.1 costs $2.00 and $8.00. Per solved SWE-bench Verified task, Devstral 2 2512 costs $1.19 and GPT-4.1 $0.37.
Which has the bigger context window, Devstral 2 2512 or GPT-4.1?
GPT-4.1, with 1.05M tokens against 262K.
How many benchmarks is the Devstral 2 2512 vs GPT-4.1 comparison based on?
3 benchmarks have a result for both models as of October 21, 2026. Devstral 2 2512's coding index rests on 1 independent benchmark and GPT-4.1's on 0. Where only one model has a result, the page says so instead of naming a winner.