Devstral 2 2512 vs GPT-4
Devstral 2 2512 leads GPT-4 on the Artificial Analysis Intelligence Index, 8.6 to 6.7, and Devstral 2 2512 costs 30.0x less per output token ($2.00 vs $60.00 per million). Updated October 22, 2026; 3 benchmarks have a result for both.
Devstral 2 2512 or GPT-4: 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 30.0x cheaper ($2.00 vs $60.00 per million) and a larger context window (262K against 8K tokens).
- Shared results
- 3
- Devstral 2 2512 only
- 3
- GPT-4 only
- 42
- 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
Devstral 2 2512
Devstral 2 2512 leads, 8.6 against 6.7.
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
Long documents
Listed context window
Devstral 2 2512
Devstral 2 2512 takes 262K tokens against 8K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Devstral 2 2512
Devstral 2 2512 $0.0014, GPT-4 $0.060 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only Devstral 2 2512 has the number, $1.19.
Directional only
At a glance
- Coding index
- Devstral 2 2512 not benchmarked, GPT-4 not benchmarked.
- Widest category
- Coding, 24.8 points apart on average over 2 shared benchmarks.
- Widest single gap
- SWE-bench Verified: Devstral 2 2512 53.8% against GPT-4 22.4%, 31.4 points apart on a 0 to 100 scale.
- Output price
- Devstral 2 2512 $2.00 against GPT-4 $60.00 per million output tokens, 30.0x apart.
- Context window
- Devstral 2 2512 takes 262K tokens, GPT-4 8K.
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 | Ties |
|---|---|---|---|---|
| Coding | 2 | 0 | ||
| Intelligence | 1 | 0 | ||
| All shared boards | 3 | 0 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index20
| Benchmark | Devstral 2 2512 | GPT-4 | Leader |
|---|---|---|---|
| SWE-bench Verified | 53.8% | 22.4% | Devstral 2 2512 |
| SWE-bench Lite | Not measured | 39.7% | Only GPT-4 measured |
| SWE-bench (full test set) | Not measured | 12.5% | Only GPT-4 measured |
| AA Coding Index | 31.3 | 13.1 | Devstral 2 2512 |
| LiveCodeBench (Vals AI) | 51.8% | Not measured | Only Devstral 2 2512 measured |
| SWE-bench Verified (Vals AI) | 62.8% | Not measured | Only Devstral 2 2512 measured |
| LMArena Text: coding | Not measured | 1209 | Only GPT-4 measured |
| LMArena Text: software and IT | Not measured | 1195 | Only GPT-4 measured |
| Neither model has a result on the other 31 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Devstral 2 2512 | GPT-4 | Leader |
|---|---|---|---|
| AA Intelligence Index | 8.6 | 6.7 | Devstral 2 2512 |
| Epoch Capabilities Index | Not measured | 125.7 | Only GPT-4 measured |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index1 board, none shared
| Benchmark | Devstral 2 2512 | GPT-4 | Leader |
|---|---|---|---|
| AA Agentic Index | 2.3 | Not measured | Only Devstral 2 2512 measured |
| Neither model has a result on the other 40 boards here | |||
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index5 boards, none shared
| Benchmark | Devstral 2 2512 | GPT-4 | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | Not measured | 34.9% | Only GPT-4 measured |
| BIG-Bench Hard | Not measured | 75.1% | Only GPT-4 measured |
| LMArena Text: hard prompts | Not measured | 1200 | Only GPT-4 measured |
| LMArena Text: expert | Not measured | 1149 | Only GPT-4 measured |
| LMArena Text: hard prompts in English | Not measured | 1209 | Only GPT-4 measured |
| Neither model has a result on the other 17 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index3 boards, none shared
| Benchmark | Devstral 2 2512 | GPT-4 | Leader |
|---|---|---|---|
| GSM8K | Not measured | 92.0% | Only GPT-4 measured |
| LMArena Text: math | Not measured | 1230 | Only GPT-4 measured |
| LMArena Text: mathematical work | Not measured | 1228 | Only GPT-4 measured |
| Neither model has a result on the other 10 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index7 boards, none shared
| Benchmark | Devstral 2 2512 | GPT-4 | Leader |
|---|---|---|---|
| MMLU Pro (Artificial Analysis run) | Not measured | 56.2% | Only GPT-4 measured |
| MMLU | Not measured | 86.4% | Only GPT-4 measured |
| AI2 Reasoning Challenge | Not measured | 96.3% | Only GPT-4 measured |
| LMArena Text: science | Not measured | 1177 | Only GPT-4 measured |
| LMArena Text: medicine and healthcare | Not measured | 1145 | Only GPT-4 measured |
| LMArena Text: legal and government | Not measured | 1191 | Only GPT-4 measured |
| LMArena Text: business and finance | Not measured | 1145 | Only GPT-4 measured |
| Neither model has a result on the other 11 boards here | |||
Web designNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Devstral 2 2512 | GPT-4 | Leader |
|---|---|---|---|
| Neither model has a result on 9 boards here | |||
GeneralDirectional onlyPublic boards, shown as published, kept out of the index22 boards, none shared
| Benchmark | Devstral 2 2512 | GPT-4 | Leader |
|---|---|---|---|
| IFBench | Not measured | 33.2% | Only GPT-4 measured |
| DTBench | Not measured | 62.7% | Only GPT-4 measured |
| LMCA | Not measured | 17.1% | Only GPT-4 measured |
| HellaSwag | Not measured | 95.3% | Only GPT-4 measured |
| WinoGrande | Not measured | 87.5% | Only GPT-4 measured |
| LMArena Text | Not measured | 1206 | Only GPT-4 measured |
| LMArena Text: instruction following | Not measured | 1201 | Only GPT-4 measured |
| LMArena Text: creative writing | Not measured | 1191 | Only GPT-4 measured |
| LMArena Text: English | Not measured | 1220 | Only GPT-4 measured |
| LMArena Text: non-English | Not measured | 1194 | Only GPT-4 measured |
| LMArena Text: Chinese | Not measured | 1184 | Only GPT-4 measured |
| LMArena Text: German | Not measured | 1198 | Only GPT-4 measured |
| LMArena Text: Japanese | Not measured | 1137 | Only GPT-4 measured |
| LMArena Text: Korean | Not measured | 1087 | Only GPT-4 measured |
| LMArena Text: French | Not measured | 1220 | Only GPT-4 measured |
| LMArena Text: Spanish | Not measured | 1197 | Only GPT-4 measured |
| LMArena Text: Russian | Not measured | 1195 | Only GPT-4 measured |
| LMArena Text: Polish | Not measured | 1265 | Only GPT-4 measured |
| LMArena Text: multi-turn | Not measured | 1206 | Only GPT-4 measured |
| LMArena Text: longer queries | Not measured | 1190 | Only GPT-4 measured |
| LMArena Text: writing and language | Not measured | 1219 | Only GPT-4 measured |
| LMArena Text: entertainment and media | Not measured | 1184 | Only GPT-4 measured |
| Neither model has a result on the other 24 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Devstral 2 2512 | GPT-4 | 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 |
|---|---|---|
| Maker | Mistral AI | OpenAI |
| Weights | Open weights | No public weights listed |
| Released | December 9, 2025 | May 28, 2023 |
| Context window | 262K | 8K |
| Input $/Mtok | $0.40 | $30.00 |
| Output $/Mtok | $2.00 | $60.00 |
| Coding index | Not benchmarked | Not benchmarked |
| Independent measurements behind it | 1 | 0 |
| Cost per solved task | $1.19 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Devstral 2 2512 better than GPT-4 for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. Pick Devstral 2 2512 for output tokens 30.0x cheaper ($2.00 vs $60.00 per million) and a larger context window (262K against 8K tokens).
Which is cheaper, Devstral 2 2512 or GPT-4?
Devstral 2 2512 costs $0.40 per million input tokens and $2.00 per million output tokens; GPT-4 costs $30.00 and $60.00. Cost per solved task is known for Devstral 2 2512 only.
Which has the bigger context window, Devstral 2 2512 or GPT-4?
Devstral 2 2512, with 262K tokens against 8K.
How many benchmarks is the Devstral 2 2512 vs GPT-4 comparison based on?
3 benchmarks have a result for both models as of October 22, 2026. Devstral 2 2512's coding index rests on 1 independent benchmark and GPT-4's on 0. Where only one model has a result, the page says so instead of naming a winner.