Devstral 2 2512 vs Qwen3.6 35B A3B
Qwen3.6 35B A3B leads Devstral 2 2512 on the Artificial Analysis Intelligence Index, 18.2 to 8.6, and Qwen3.6 35B A3B costs 2.0x less per output token ($1.00 vs $2.00 per million). Updated October 22, 2026; 3 benchmarks have a result for both.
Devstral 2 2512 or Qwen3.6 35B A3B: which to pick
Neither model has a coding score yet, so the two cannot be ranked on coding.
Pick Devstral 2 2512 for work measured by AA Coding Index or AA Intelligence Index, where it leads.
- Shared results
- 3
- Devstral 2 2512 only
- 3
- Qwen3.6 35B A3B only
- 17
- Like-for-like categories
- 3 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
Qwen3.6 35B A3B
Qwen3.6 35B A3B leads, 18.2 against 8.6.
Same board, both measured
Agentic work
AA Agentic Index
Qwen3.6 35B A3B
Qwen3.6 35B A3B leads, 13.1 against 2.3.
Same board, both measured
Long documents
Listed context window
No call
Both take 262K tokens.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Qwen3.6 35B A3B
Devstral 2 2512 $0.0014, Qwen3.6 35B A3B $0.0006 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, Qwen3.6 35B A3B not benchmarked.
- Widest category
- Agentic work, 10.8 points apart on average over 1 shared benchmark.
- Widest single gap
- AA Agentic Index: Devstral 2 2512 2.3 against Qwen3.6 35B A3B 13.1, 10.8 points apart on a 0 to 100 scale.
- Output price
- Devstral 2 2512 $2.00 against Qwen3.6 35B A3B $1.00 per million output tokens, 2.0x apart.
- Context window
- Both take 262K tokens.
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 | Qwen3.6 35B A3B | Ties |
|---|---|---|---|---|
| Coding | 0 | 1 | ||
| Intelligence | 0 | 1 | ||
| Agentic work | 0 | 1 | ||
| All shared boards | 0 | 3 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index01
| Benchmark | Devstral 2 2512 | Qwen3.6 35B A3B | Leader |
|---|---|---|---|
| SWE-bench Verified | 53.8% | Not measured | Only Devstral 2 2512 measured |
| AA Coding Index | 31.3 | 41.9 | Qwen3.6 35B A3B |
| 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 |
| SciCode | Not measured | 36.6% | Only Qwen3.6 35B A3B measured |
| Terminal-Bench Hard (Artificial Analysis run) | Not measured | 34.8% | Only Qwen3.6 35B A3B measured |
| Terminal-Bench | Not measured | 24.6% | Only Qwen3.6 35B A3B measured |
| WeirdML | Not measured | 34.5% | Only Qwen3.6 35B A3B measured |
| Neither model has a result on the other 31 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Devstral 2 2512 | Qwen3.6 35B A3B | Leader |
|---|---|---|---|
| AA Intelligence Index | 8.6 | 18.2 | Qwen3.6 35B A3B |
| Epoch Capabilities Index | Not measured | 143.7 | Only Qwen3.6 35B A3B measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Devstral 2 2512 | Qwen3.6 35B A3B | Leader |
|---|---|---|---|
| AA Agentic Index | 2.3 | 13.1 | Qwen3.6 35B A3B |
| τ²-Bench Telecom | Not measured | 95.3% | Only Qwen3.6 35B A3B measured |
| Neither model has a result on the other 39 boards here | |||
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index4 boards, none shared
| Benchmark | Devstral 2 2512 | Qwen3.6 35B A3B | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | Not measured | 84.1% | Only Qwen3.6 35B A3B measured |
| Humanity's Last Exam (Artificial Analysis run) | Not measured | 22.2% | Only Qwen3.6 35B A3B measured |
| GPQA Diamond (Epoch AI run) | Not measured | 84.8% | Only Qwen3.6 35B A3B measured |
| Mystery game puzzles | Not measured | 22.0% | Only Qwen3.6 35B A3B measured |
| Neither model has a result on the other 18 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | Devstral 2 2512 | Qwen3.6 35B A3B | Leader |
|---|---|---|---|
| OTIS Mock AIME | Not measured | 86.7% | Only Qwen3.6 35B A3B measured |
| FrontierMath, tiers 1 to 3 | Not measured | 20.4% | Only Qwen3.6 35B A3B measured |
| Neither model has a result on the other 11 boards here | |||
KnowledgeNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Devstral 2 2512 | Qwen3.6 35B A3B | Leader |
|---|---|---|---|
| Neither model has a result on 18 boards here | |||
Web designNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Devstral 2 2512 | Qwen3.6 35B A3B | Leader |
|---|---|---|---|
| Neither model has a result on 9 boards here | |||
GeneralDirectional onlyPublic boards, shown as published, kept out of the index5 boards, none shared
| Benchmark | Devstral 2 2512 | Qwen3.6 35B A3B | Leader |
|---|---|---|---|
| IFBench | Not measured | 64.4% | Only Qwen3.6 35B A3B measured |
| Long context reasoning (AA-LCR) | Not measured | 71.7% | Only Qwen3.6 35B A3B measured |
| DTBench | Not measured | 73.9% | Only Qwen3.6 35B A3B measured |
| LMCA | Not measured | 29.7% | Only Qwen3.6 35B A3B measured |
| Surface Evolver Bench | Not measured | 44.4% | Only Qwen3.6 35B A3B measured |
| Neither model has a result on the other 41 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Devstral 2 2512 | Qwen3.6 35B A3B | 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
Qwen3.6 35B A3B costs less
Repository review
50K input + 3K output tokens
Qwen3.6 35B A3B costs less
Agent loop
200K input + 10K output tokens
Qwen3.6 35B A3B 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 | Qwen3.6 35B A3B |
|---|---|---|
| Maker | Mistral AI | Qwen |
| Weights | Open weights | Open weights |
| Released | December 9, 2025 | April 27, 2026 |
| Context window | 262K | 262K |
| Input $/Mtok | $0.40 | $0.10 |
| Output $/Mtok | $2.00 | $1.00 |
| 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 Qwen3.6 35B A3B for coding?
Neither model has a coding score yet, so the two cannot be ranked on coding. Pick Devstral 2 2512 for work measured by AA Coding Index or AA Intelligence Index, where it leads.
Which is cheaper, Devstral 2 2512 or Qwen3.6 35B A3B?
Devstral 2 2512 costs $0.40 per million input tokens and $2.00 per million output tokens; Qwen3.6 35B A3B costs $0.10 and $1.00. Cost per solved task is known for Devstral 2 2512 only.
Which has the bigger context window, Devstral 2 2512 or Qwen3.6 35B A3B?
Neither: both take 262K tokens.
How many benchmarks is the Devstral 2 2512 vs Qwen3.6 35B A3B 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 Qwen3.6 35B A3B's on 0. Where only one model has a result, the page says so instead of naming a winner.