Trinity Large Thinking vs Devstral 2 2512
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 5.4 for Trinity Large Thinking.
Pick Trinity Large Thinking when output tokens cost 2.5x less, or when the benchmark that matters is AA Intelligence Index, where it leads.
- Shared results
- 3
- Trinity Large Thinking only
- 49
- Devstral 2 2512 only
- 3
- Like-for-like categories
- 3 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 5.4.
One benchmark behind each score
Intelligence
AA Intelligence Index
Trinity Large Thinking
Trinity Large Thinking leads, 10.8 against 8.6.
Same board, both measured
Agentic work
AA Agentic Index
Devstral 2 2512
Devstral 2 2512 leads, 2.3 against 1.0.
Same board, both measured
Web design
Design Arena: websites
No call
Only Trinity Large Thinking has a Design Arena: websites result, 1137.
Directional only
Long documents
Listed context window
No call
Both take 262K tokens.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Trinity Large Thinking
Trinity Large Thinking $0.0006, Devstral 2 2512 $0.0014 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
- Trinity Large Thinking 5.4 from 1 benchmark, Devstral 2 2512 53.8 from 1 benchmark.
- Widest category
- Coding, 5.5 points apart on average over 1 shared benchmark.
- Widest single gap
- AA Coding Index: Trinity Large Thinking 25.8 against Devstral 2 2512 31.3, 5.5 points apart on a 0 to 100 scale.
- Output price
- Trinity Large Thinking $0.80 against Devstral 2 2512 $2.00 per million output tokens, 2.5x 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 | Trinity Large Thinking | Bars | Devstral 2 2512 | Ties |
|---|---|---|---|---|
| Coding | 0 | 1 | ||
| Intelligence | 1 | 0 | ||
| Agentic work | 0 | 1 | ||
| All shared boards | 1 | 2 |
CodingLike-for-likeFour of these boards feed the Vibe Coding Index01
| Benchmark | Trinity Large Thinking | Devstral 2 2512 | Leader |
|---|---|---|---|
| SWE-bench Verified | Not measured | 53.8% | Only Devstral 2 2512 measured |
| LMArena WebDev | 1238 | Not measured | Only Trinity Large Thinking measured |
| AA Coding Index | 25.8 | 31.3 | Devstral 2 2512 |
| LiveCodeBench (Vals AI) | Not measured | 51.8% | Only Devstral 2 2512 measured |
| SWE-bench Verified (Vals AI) | Not measured | 62.8% | Only Devstral 2 2512 measured |
| SciCode | 40.6% | Not measured | Only Trinity Large Thinking measured |
| Terminal-Bench Hard (Artificial Analysis run) | 22.7% | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: coding | 1357 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: software and IT | 1362 | Not measured | Only Trinity Large Thinking measured |
| LMArena WebDev: React | 1231 | Not measured | Only Trinity Large Thinking measured |
| LMArena WebDev: HTML | 1240 | Not measured | Only Trinity Large Thinking measured |
| Neither model has a result on the other 27 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index10
| Benchmark | Trinity Large Thinking | Devstral 2 2512 | Leader |
|---|---|---|---|
| AA Intelligence Index | 10.8 | 8.6 | Trinity Large Thinking |
| Neither model has a result on the other 1 board here | |||
Agentic workLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | Trinity Large Thinking | Devstral 2 2512 | Leader |
|---|---|---|---|
| AA Agentic Index | 1.0 | 2.3 | Devstral 2 2512 |
| τ²-Bench Telecom | 90.1% | Not measured | Only Trinity Large Thinking measured |
| Neither model has a result on the other 37 boards here | |||
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index5 boards, none shared
| Benchmark | Trinity Large Thinking | Devstral 2 2512 | Leader |
|---|---|---|---|
| GPQA Diamond (Artificial Analysis run) | 75.2% | Not measured | Only Trinity Large Thinking measured |
| Humanity's Last Exam (Artificial Analysis run) | 15.8% | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: hard prompts | 1342 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: expert | 1359 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: hard prompts in English | 1356 | Not measured | Only Trinity Large Thinking measured |
| Neither model has a result on the other 16 boards here | |||
MathDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | Trinity Large Thinking | Devstral 2 2512 | Leader |
|---|---|---|---|
| LMArena Text: math | 1365 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: mathematical work | 1350 | Not measured | Only Trinity Large Thinking measured |
| Neither model has a result on the other 11 boards here | |||
KnowledgeDirectional onlyPublic boards, shown as published, kept out of the index6 boards, none shared
| Benchmark | Trinity Large Thinking | Devstral 2 2512 | Leader |
|---|---|---|---|
| CaseLaw (Vals AI) | 57.9% | Not measured | Only Trinity Large Thinking measured |
| CorpFin (Vals AI) | 54.7% | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: science | 1355 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: medicine and healthcare | 1349 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: legal and government | 1345 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: business and finance | 1329 | Not measured | Only Trinity Large Thinking measured |
| Neither model has a result on the other 11 boards here | |||
Web designDirectional onlyPublic boards, shown as published, kept out of the index8 boards, none shared
| Benchmark | Trinity Large Thinking | Devstral 2 2512 | Leader |
|---|---|---|---|
| Design Arena: websites | 1137 | Not measured | Only Trinity Large Thinking measured |
| Design Arena: UI components | 1045 | Not measured | Only Trinity Large Thinking measured |
| Design Arena: game dev | 1079 | Not measured | Only Trinity Large Thinking measured |
| Design Arena: data viz | 1098 | Not measured | Only Trinity Large Thinking measured |
| Design Arena: 3D | 1082 | Not measured | Only Trinity Large Thinking measured |
| Design Arena: all code categories | 1117 | Not measured | Only Trinity Large Thinking measured |
| Design Arena: SVG | 1016 | Not measured | Only Trinity Large Thinking measured |
| Design Arena: ASCII art | 1041 | Not measured | Only Trinity Large Thinking measured |
| Neither model has a result on the other 1 board here | |||
GeneralDirectional onlyPublic boards, shown as published, kept out of the index20 boards, none shared
| Benchmark | Trinity Large Thinking | Devstral 2 2512 | Leader |
|---|---|---|---|
| IFBench | 56.3% | Not measured | Only Trinity Large Thinking measured |
| Long context reasoning (AA-LCR) | 38.0% | Not measured | Only Trinity Large Thinking measured |
| Surface Evolver Bench | 15.6% | Not measured | Only Trinity Large Thinking measured |
| LMArena Text | 1340 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: instruction following | 1326 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: creative writing | 1308 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: English | 1356 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: non-English | 1325 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: Chinese | 1371 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: German | 1356 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: Japanese | 1312 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: Korean | 1306 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: French | 1366 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: Spanish | 1327 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: Russian | 1338 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: Polish | 1312 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: multi-turn | 1320 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: longer queries | 1318 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: writing and language | 1317 | Not measured | Only Trinity Large Thinking measured |
| LMArena Text: entertainment and media | 1296 | Not measured | Only Trinity Large Thinking measured |
| Neither model has a result on the other 26 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Trinity Large Thinking | Devstral 2 2512 | 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
Trinity Large Thinking costs less
Repository review
50K input + 3K output tokens
Trinity Large Thinking costs less
Agent loop
200K input + 10K output tokens
Trinity Large Thinking 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 | Arcee AI Trinity Large Thinking | Mistral Devstral 2 2512 |
|---|---|---|
| Maker | Arcee AI | Mistral AI |
| Weights | Open weights | Open weights |
| Released | April 1, 2026 | December 9, 2025 |
| Context window | 262K | 262K |
| Input $/Mtok | $0.25 | $0.40 |
| Output $/Mtok | $0.80 | $2.00 |
| Coding index | 5.4 | 53.8 |
| Independent measurements behind it | 1 | 1 |
| Cost per solved task | Not computable | $1.19 |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Trinity Large Thinking better than Devstral 2 2512 for coding?
Pick Devstral 2 2512 for coding: it scores 53.8 on the Vibe Coding Index against 5.4 for Trinity Large Thinking. Pick Trinity Large Thinking when output tokens cost 2.5x less, or when the benchmark that matters is AA Intelligence Index, where it leads.
Which is cheaper, Trinity Large Thinking or Devstral 2 2512?
Trinity Large Thinking costs $0.25 per million input tokens and $0.80 per million output tokens; Devstral 2 2512 costs $0.40 and $2.00. Cost per solved task is known for Devstral 2 2512 only.
Which has the bigger context window, Trinity Large Thinking or Devstral 2 2512?
Neither: both take 262K tokens.
How many benchmarks is the Trinity Large Thinking vs Devstral 2 2512 comparison based on?
3 benchmarks have a result for both models as of October 21, 2026. Trinity Large Thinking's coding index rests on 1 independent benchmark and Devstral 2 2512's on 1. Where only one model has a result, the page says so instead of naming a winner.