Devstral 2 2512 vs Gemini 3.5 Flash
Updated October 21, 2026. 3 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Pick Devstral 2 2512 for coding: it scores 53.8 on the Vibe Coding Index against 42.8 for Gemini 3.5 Flash.
Pick Gemini 3.5 Flash when the context window is larger, 1.05M against 262K tokens, or when what matters is LiveCodeBench (Vals AI) or SWE-bench Verified (Vals AI), where it leads.
- Shared results
- 3
- Devstral 2 2512 only
- 3
- Gemini 3.5 Flash only
- 107
- Like-for-like categories
- 2 of 9
Shared results can support a head-to-head reading. Results only one model has describe coverage, not superiority.
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, four public coding benchmarks
Devstral 2 2512
Devstral 2 2512 scores 53.8 against 42.8.
One benchmark behind each score
Intelligence
AA Intelligence Index
Gemini 3.5 Flash
Gemini 3.5 Flash leads, 33.6 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 Gemini 3.5 Flash has a GPQA Diamond result, 92.7%.
Directional only
Knowledge
MMLU Pro
No call
Only Gemini 3.5 Flash has a MMLU Pro result, 89.5%.
Directional only
Web design
Design Arena: websites
No call
Only Gemini 3.5 Flash has a Design Arena: websites result, 1268.
Directional only
Long documents
Listed context window
Gemini 3.5 Flash
Gemini 3.5 Flash takes 1.05M tokens against 262K.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Devstral 2 2512
Devstral 2 2512 $0.0014, Gemini 3.5 Flash $0.0060 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
Devstral 2 2512
Devstral 2 2512 $1.19, Gemini 3.5 Flash $1.21.
Published attempt cost and pass rate
At a glance
- Coding index
- Devstral 2 2512 53.8 from 1 benchmark, Gemini 3.5 Flash 42.8 from 1 benchmark.
- Widest category
- Coding, 25.9 points apart on average over 2 shared benchmarks.
- Widest single gap
- LiveCodeBench (Vals AI): Devstral 2 2512 51.8% against Gemini 3.5 Flash 87.6%, 35.8 points apart on a 0 to 100 scale.
- Output price
- Devstral 2 2512 $2.00 against Gemini 3.5 Flash $9.00 per million output tokens, 4.5x apart.
- Context window
- Devstral 2 2512 takes 262K tokens, Gemini 3.5 Flash 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 | Gemini 3.5 Flash | Ties |
|---|---|---|---|---|
| Coding | 0 | 2 | ||
| Intelligence | 0 | 1 | ||
| All shared boards | 0 | 3 |
CodingLike-for-likeFour of these boards feed the Vibe Coding Index02
| Benchmark | Devstral 2 2512 | Gemini 3.5 Flash | Leader |
|---|---|---|---|
| SWE-bench Verified | 53.8% | Not measured | Only Devstral 2 2512 measured |
| LMArena WebDev | Not measured | 1499 | Only Gemini 3.5 Flash measured |
| AA Coding Index | 31.3 | Not measured | Only Devstral 2 2512 measured |
| LiveCodeBench (Vals AI) | 51.8% | 87.6% | Gemini 3.5 Flash |
| SWE-bench Verified (Vals AI) | 62.8% | 78.8% | Gemini 3.5 Flash |
| Terminal-Bench 4 (Vals AI) | Not measured | 6.1% | Only Gemini 3.5 Flash measured |
| Vibe Code Bench (Vals AI) | Not measured | 48.7% | Only Gemini 3.5 Flash measured |
| Code migration (Vals AI) | Not measured | 26.8% | Only Gemini 3.5 Flash measured |
| ProgramBench (Vals AI) | Not measured | 0.0% | Only Gemini 3.5 Flash measured |
| Terminal-Bench 2.1 (Vals AI) | Not measured | 74.2% | Only Gemini 3.5 Flash measured |
| SkillsBench (Vals AI) | Not measured | 52.7% | Only Gemini 3.5 Flash measured |
| SciCode | Not measured | 53.9% | Only Gemini 3.5 Flash measured |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 46.2% | Only Gemini 3.5 Flash measured |
| WeirdML | Not measured | 62.6% | Only Gemini 3.5 Flash measured |
| ALE-Bench | Not measured | 911.0 | Only Gemini 3.5 Flash measured |
| LMArena Text: coding | Not measured | 1493 | Only Gemini 3.5 Flash measured |
| LMArena Text: software and IT | Not measured | 1494 | Only Gemini 3.5 Flash measured |
| LMArena WebDev: React | Not measured | 1495 | Only Gemini 3.5 Flash measured |
| LMArena WebDev: HTML | Not measured | 1522 | Only Gemini 3.5 Flash measured |
| LMArena WebDev: from an image | Not measured | 1504 | Only Gemini 3.5 Flash measured |
| Neither model has a result on the other 18 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Devstral 2 2512 | Gemini 3.5 Flash | Leader |
|---|---|---|---|
| AA Intelligence Index | 8.6 | 33.6 | Gemini 3.5 Flash |
| Epoch Capabilities Index | Not measured | 154.9 | Only Gemini 3.5 Flash measured |
Agentic workDirectional onlyPublic boards, shown as published, kept out of the index17 boards, none shared
| Benchmark | Devstral 2 2512 | Gemini 3.5 Flash | Leader |
|---|---|---|---|
| AA Agentic Index | 2.3 | Not measured | Only Devstral 2 2512 measured |
| Design Arena: agents building web apps | Not measured | 1202 | Only Gemini 3.5 Flash measured |
| Design Arena: agents building full-stack apps | Not measured | 1197 | Only Gemini 3.5 Flash measured |
| Design Arena: agents building mobile apps | Not measured | 1175 | Only Gemini 3.5 Flash measured |
| Design Arena: agents building Android apps | Not measured | 1190 | Only Gemini 3.5 Flash measured |
| Design Arena: agents building Godot games | Not measured | 1098 | Only Gemini 3.5 Flash measured |
| Design Arena: agents building games | Not measured | 1180 | Only Gemini 3.5 Flash measured |
| Design Arena: HTML slides | Not measured | 1142 | Only Gemini 3.5 Flash measured |
| Design Arena: PowerPoint slides | Not measured | 1178 | Only Gemini 3.5 Flash measured |
| Finance Agent (Vals AI) | Not measured | 57.9% | Only Gemini 3.5 Flash measured |
| Tax Agent Bench (Vals AI) | Not measured | 21.6% | Only Gemini 3.5 Flash measured |
| Legal Research Bench (Vals AI) | Not measured | 30.8% | Only Gemini 3.5 Flash measured |
| Excel Modeling Benchmark (Vals AI) | Not measured | 63.5% | Only Gemini 3.5 Flash measured |
| Terminal-Bench-Science (Vals AI) | Not measured | 5.7% | Only Gemini 3.5 Flash measured |
| τ²-Bench Telecom | Not measured | 95.6% | Only Gemini 3.5 Flash measured |
| APEX-Agents | Not measured | 27.5% | Only Gemini 3.5 Flash measured |
| Vending-Bench 2 | Not measured | $5,396 | Only Gemini 3.5 Flash measured |
| Neither model has a result on the other 22 boards here | |||
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index16 boards, none shared
| Benchmark | Devstral 2 2512 | Gemini 3.5 Flash | Leader |
|---|---|---|---|
| GPQA Diamond | Not measured | 92.7% | Only Gemini 3.5 Flash measured |
| MMMU Pro (Vals AI) | Not measured | 88.3% | Only Gemini 3.5 Flash measured |
| GPQA Diamond (Artificial Analysis run) | Not measured | 92.2% | Only Gemini 3.5 Flash measured |
| Humanity's Last Exam (Artificial Analysis run) | Not measured | 42.7% | Only Gemini 3.5 Flash measured |
| GPQA Diamond (Epoch AI run) | Not measured | 92.8% | Only Gemini 3.5 Flash measured |
| ARC-AGI | Not measured | 92.5% | Only Gemini 3.5 Flash measured |
| ARC-AGI-2 | Not measured | 72.1% | Only Gemini 3.5 Flash measured |
| SimpleBench | Not measured | 76.7% | Only Gemini 3.5 Flash measured |
| Chess puzzles | Not measured | 45.0% | Only Gemini 3.5 Flash measured |
| Mystery game puzzles | Not measured | 32.0% | Only Gemini 3.5 Flash measured |
| EBR-bench | Not measured | 4.8% | Only Gemini 3.5 Flash measured |
| LMArena Text: hard prompts | Not measured | 1488 | Only Gemini 3.5 Flash measured |
| LMArena Text: expert | Not measured | 1495 | Only Gemini 3.5 Flash measured |
| LMArena Text: hard prompts in English | Not measured | 1483 | Only Gemini 3.5 Flash measured |
| LMArena Vision: diagrams | Not measured | 1323 | Only Gemini 3.5 Flash measured |
| LMArena Vision: homework | Not measured | 1330 | Only Gemini 3.5 Flash measured |
| Neither model has a result on the other 5 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index6 boards, none shared
| Benchmark | Devstral 2 2512 | Gemini 3.5 Flash | Leader |
|---|---|---|---|
| OTIS Mock AIME | Not measured | 95.6% | Only Gemini 3.5 Flash measured |
| FrontierMath, tiers 1 to 3 | Not measured | 62.8% | Only Gemini 3.5 Flash measured |
| FrontierMath, tier 4 | Not measured | 26.8% | Only Gemini 3.5 Flash measured |
| ProofBench | Not measured | 31.0% | Only Gemini 3.5 Flash measured |
| LMArena Text: math | Not measured | 1504 | Only Gemini 3.5 Flash measured |
| LMArena Text: mathematical work | Not measured | 1495 | Only Gemini 3.5 Flash measured |
| Neither model has a result on the other 7 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index11 boards, none shared
| Benchmark | Devstral 2 2512 | Gemini 3.5 Flash | Leader |
|---|---|---|---|
| MMLU Pro | Not measured | 89.5% | Only Gemini 3.5 Flash measured |
| LegalBench (Vals AI) | Not measured | 83.6% | Only Gemini 3.5 Flash measured |
| CorpFin (Vals AI) | Not measured | 64.7% | Only Gemini 3.5 Flash measured |
| TaxEval (Vals AI) | Not measured | 74.4% | Only Gemini 3.5 Flash measured |
| MortgageTax (Vals AI) | Not measured | 68.1% | Only Gemini 3.5 Flash measured |
| MedCode (Vals AI) | Not measured | 55.8% | Only Gemini 3.5 Flash measured |
| SimpleQA Verified | Not measured | 66.2% | Only Gemini 3.5 Flash measured |
| LMArena Text: science | Not measured | 1495 | Only Gemini 3.5 Flash measured |
| LMArena Text: medicine and healthcare | Not measured | 1483 | Only Gemini 3.5 Flash measured |
| LMArena Text: legal and government | Not measured | 1488 | Only Gemini 3.5 Flash measured |
| LMArena Text: business and finance | Not measured | 1467 | Only Gemini 3.5 Flash measured |
| Neither model has a result on the other 6 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | Devstral 2 2512 | Gemini 3.5 Flash | Leader |
|---|---|---|---|
| Design Arena: websites | Not measured | 1268 | Only Gemini 3.5 Flash measured |
| Design Arena: UI components | Not measured | 1272 | Only Gemini 3.5 Flash measured |
| Design Arena: game dev | Not measured | 1263 | Only Gemini 3.5 Flash measured |
| Design Arena: data viz | Not measured | 1236 | Only Gemini 3.5 Flash measured |
| Design Arena: 3D | Not measured | 1244 | Only Gemini 3.5 Flash measured |
| Design Arena: all code categories | Not measured | 1266 | Only Gemini 3.5 Flash measured |
| Design Arena: SVG | Not measured | 1254 | Only Gemini 3.5 Flash measured |
| Design Arena: ASCII art | Not measured | 1253 | Only Gemini 3.5 Flash measured |
| Neither model has a result on the other 1 board here | |||
GeneralDirectional onlyPublic boards, shown as published, kept out of the index33 boards, none shared
| Benchmark | Devstral 2 2512 | Gemini 3.5 Flash | Leader |
|---|---|---|---|
| Public Benefits Bench (Vals AI) | Not measured | 59.5% | Only Gemini 3.5 Flash measured |
| SAGE (Vals AI) | Not measured | 49.9% | Only Gemini 3.5 Flash measured |
| MedScribe (Vals AI) | Not measured | 76.6% | Only Gemini 3.5 Flash measured |
| Vals Multimodal Index | Not measured | 62.9% | Only Gemini 3.5 Flash measured |
| IFBench | Not measured | 76.3% | Only Gemini 3.5 Flash measured |
| Long context reasoning (AA-LCR) | Not measured | 74.3% | Only Gemini 3.5 Flash measured |
| DTBench | Not measured | 94.7% | Only Gemini 3.5 Flash measured |
| LMCA | Not measured | 47.1% | Only Gemini 3.5 Flash measured |
| Surface Evolver Bench | Not measured | 58.1% | Only Gemini 3.5 Flash measured |
| LMArena Text | Not measured | 1482 | Only Gemini 3.5 Flash measured |
| LMArena Text: instruction following | Not measured | 1467 | Only Gemini 3.5 Flash measured |
| LMArena Text: creative writing | Not measured | 1470 | Only Gemini 3.5 Flash measured |
| LMArena Vision | Not measured | 1308 | Only Gemini 3.5 Flash measured |
| LMArena Text: English | Not measured | 1481 | Only Gemini 3.5 Flash measured |
| LMArena Text: non-English | Not measured | 1475 | Only Gemini 3.5 Flash measured |
| LMArena Text: Chinese | Not measured | 1527 | Only Gemini 3.5 Flash measured |
| LMArena Text: German | Not measured | 1494 | Only Gemini 3.5 Flash measured |
| LMArena Text: Japanese | Not measured | 1486 | Only Gemini 3.5 Flash measured |
| LMArena Text: Korean | Not measured | 1448 | Only Gemini 3.5 Flash measured |
| LMArena Text: French | Not measured | 1490 | Only Gemini 3.5 Flash measured |
| LMArena Text: Spanish | Not measured | 1480 | Only Gemini 3.5 Flash measured |
| LMArena Text: Russian | Not measured | 1493 | Only Gemini 3.5 Flash measured |
| LMArena Text: Polish | Not measured | 1516 | Only Gemini 3.5 Flash measured |
| LMArena Text: multi-turn | Not measured | 1481 | Only Gemini 3.5 Flash measured |
| LMArena Text: longer queries | Not measured | 1483 | Only Gemini 3.5 Flash measured |
| LMArena Text: writing and language | Not measured | 1478 | Only Gemini 3.5 Flash measured |
| LMArena Text: entertainment and media | Not measured | 1459 | Only Gemini 3.5 Flash measured |
| LMArena Vision: English | Not measured | 1305 | Only Gemini 3.5 Flash measured |
| LMArena Vision: Chinese | Not measured | 1377 | Only Gemini 3.5 Flash measured |
| LMArena Vision: reading text | Not measured | 1317 | Only Gemini 3.5 Flash measured |
| LMArena Vision: humor | Not measured | 1321 | Only Gemini 3.5 Flash measured |
| LMArena Vision: creative writing | Not measured | 1301 | Only Gemini 3.5 Flash measured |
| LMArena Document | Not measured | 1463 | Only Gemini 3.5 Flash measured |
| Neither model has a result on the other 13 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Devstral 2 2512 | Gemini 3.5 Flash | 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 | Google Gemini 3.5 Flash |
|---|---|---|
| Maker | Mistral AI | |
| Weights | Open weights | No public weights listed |
| Released | December 9, 2025 | May 19, 2026 |
| Context window | 262K | 1.05M |
| Input $/Mtok | $0.40 | $1.50 |
| Output $/Mtok | $2.00 | $9.00 |
| Coding index | 53.8 | 42.8 |
| Independent measurements behind it | 1 | 1 |
| Cost per solved task | $1.19 | $1.21 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Devstral 2 2512 better than Gemini 3.5 Flash for coding?
Pick Devstral 2 2512 for coding: it scores 53.8 on the Vibe Coding Index against 42.8 for Gemini 3.5 Flash. Pick Gemini 3.5 Flash when the context window is larger, 1.05M against 262K tokens, or when what matters is LiveCodeBench (Vals AI) or SWE-bench Verified (Vals AI), where it leads.
Which is cheaper, Devstral 2 2512 or Gemini 3.5 Flash?
Devstral 2 2512 costs $0.40 per million input tokens and $2.00 per million output tokens; Gemini 3.5 Flash costs $1.50 and $9.00. Per solved SWE-bench Verified task, Devstral 2 2512 costs $1.19 and Gemini 3.5 Flash $1.21.
Which has the bigger context window, Devstral 2 2512 or Gemini 3.5 Flash?
Gemini 3.5 Flash, with 1.05M tokens against 262K.
How many benchmarks is the Devstral 2 2512 vs Gemini 3.5 Flash 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 Gemini 3.5 Flash's on 1. Where only one model has a result, the page says so instead of naming a winner.