Trinity-Large-Thinking model card (Hugging Face) Apr 1, 2026
- IFBench 52.3
- GPQA-Diamond 76.3
- Mean of Tau2-Airline 88.0 / Tau2-Telecom 94.7 91.3
- PinchBench 91.9
- AIME25 96.3
- BCFLv4 70.1
- MMLU-Pro 83.4
- SWE-bench Verified (mini-swe-agent-v2) 63.2
Trinity Large Thinking
Trinity Large Thinking places in the top three on PinchBench.
29
Benchmarks with a score
0
Ranked first, out of five or more models
1
In the top three
Arcee AI Apr 2026 Trinity Large Thinking
Open weights
142.1
#161 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
- IFBench 52.3% #64 /80
- BFCL V4 70.1% #9 /30
- CorpFin v2 54.7% #86 /119
Where it ranks
Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.
Knowledge and language
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| MMLU-Pro | 83.4% | #49 of 167 | Lab model card | |
| LMArena (Elo) | 1,367 | #162 of 335 | Official leaderboard | |
| BullshitBench | 20% | #72 of 107 | Official leaderboard | |
| BullshitBench V2 | 17% | #100 of 129 | Official leaderboard | |
| IFBench | 52.3% | #64 of 80 | Lab model card | |
| Extended NYT Connections | 16.5% | #91 of 103 | Official leaderboard | |
| DesignArena ASCII Art | 1,036 | #74 of 77 | Official leaderboard |
Math and science
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| AIME 2025 | 96.3% | #19 of 143 | Lab model card | |
| GPQA Diamond | 76.3% | #139 of 302 | Lab model card | |
| Surface Evolver Bench | 15.6% | #24 of 25 | Official leaderboard |
Coding
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| SWE-bench Verified | 63.2% | #103 of 155 | Lab model card | |
| DesignArena Website | 1,130 | #108 of 151 | Official leaderboard | |
| DesignArena 3D Design | 1,064 | #103 of 136 | Official leaderboard | |
| DesignArena Code | 1,108 | #109 of 143 | Official leaderboard | |
| DesignArena Game Dev | 1,069 | #111 of 143 | Official leaderboard | |
| DesignArena UI Component | 1,036 | #116 of 139 | Official leaderboard | |
| DesignArena Data Visualization | 1,096 | #103 of 118 | Official leaderboard | |
| DesignArena SVG | 1,007 | #92 of 103 | Official leaderboard | |
| Code Arena | 1,236 | #119 of 128 | Official leaderboard | |
| WebDev Arena | 1,237 | #98 of 105 | arena.ai, via Epoch AI | |
| OpenHands Index | 32.1% | #32 of 33 | Official leaderboard |
Agents and tools
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| tau2-bench | 91.3% | #14 of 79 | Lab model card | |
| PinchBench | 91.9% | #3 of 14 | Lab model card | |
| BFCL V4 | 70.1% | #9 of 30 | Lab model card | |
| PACT | 1,526 | #12 of 26 | Official leaderboard | |
| PinchBench 2.0 | 65.7% | #38 of 57 | Official leaderboard | |
| SnakeBench | 20 | #140 of 188 | Official leaderboard |
Professional work
| Benchmark | Score | Rank | Source | |
|---|---|---|---|---|
| CaseLaw v2 | 57.9% | #29 of 53 | Official leaderboard | |
| CorpFin v2 | 54.7% | #86 of 119 | Official leaderboard |
Its lab's numbers put it at 142.1, number 161 of 404
The Self-Reported Index fits one score to the 8 capability results Arcee AI reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 138.0 to 146.3. How we compute it.
- AIME 2025 96.3
- PinchBench 91.9
- tau2-bench 91.3
- MMLU-Pro 83.4
- GPQA Diamond 76.3
- BFCL V4 70.1
- SWE-bench Verified 63.2
- IFBench 52.3
Its launch post
The benchmarks the lab chose to headline, with the scores it gave.