Trinity Large
Best configuration ranks #318 of 610 on the BenchLeader Index at 48.7 ±5.0 (thinking reasoning effort). Last measured 8 Sept 2026. Released 1 Apr 2026.
- Blended price
- $0.413/M
- $0.250 in · $0.900 out
- Output speed
- 321 tok/s
- First answer
- 7.20 s
- first token 0.21 s
- Context
- 512k
- Overall index49
- Reasoning55
- Coding37
- Agents & tools53
- Knowledge44
- Instruction following56
- Human preference55
- Long context45
- Composite43
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Default” means the publisher did not say which setting was used.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| thinkingbest | 48.7 | #318 | 321 tok/s | 7.20 s | $0.0004 | Agents & tools 53 · Coding 37 · Composite 43 · Human preference 55 · Instruction following 56 · Knowledge 44 · Long context 45 · Reasoning 55 |
| default | – | – | 105 tok/s | 0.21 s | – | Coding 58 · Human preference 56 · Reasoning 57 |
Benchmark results
One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | thinking | default | Source | Trend |
|---|---|---|---|---|
| LMArena Hard Prompts | 1386#159 | 1404#149 | LMArena | |
| GPQA Diamond (AA)not in index | 75.2%#242 | – | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 15.8%#190 | – | Artificial Analysis |
Coding
| Benchmark | thinking | default | Source | Trend |
|---|---|---|---|---|
| SciCode | 36.1%#125 | – | SciCode | |
| LMArena Coding | 1414#162 | 1442#136 | LMArena | |
| LMArena WebDev | 1237#110 | – | LMArena | |
| SciCode (AA)not in index | 40.6%#112 | – | Artificial Analysis |
Agents & tools
| Benchmark | thinking | default | Source | Trend |
|---|---|---|---|---|
| Terminal-Bench Hard | 22.7%#146 | – | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 90.1%#66 | – | Artificial Analysis |
Knowledge
| Benchmark | thinking | default | Source | Trend |
|---|---|---|---|---|
| AA-Omniscience | -44.1#281 | – | Artificial Analysis | |
| CorpFin | 54.7%#90 | – | Vals AI |
Instruction following
| Benchmark | thinking | default | Source | Trend |
|---|---|---|---|---|
| IFBench | 56.3%#141 | – | Artificial Analysis |
Human preference
| Benchmark | thinking | default | Source | Trend |
|---|---|---|---|---|
| LMArena Text | 1369#161 | 1379#153 | LMArena |
Long context
| Benchmark | thinking | default | Source | Trend |
|---|---|---|---|---|
| AA-LCR | 38.0%#294 | – | Artificial Analysis |
Composite
| Benchmark | thinking | default | Source | Trend |
|---|---|---|---|---|
| AA Intelligence Index | 10.9#300 | – | Artificial Analysis |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| Arcee AI | 105 tok/s | 0.21 s | $0.250 | $0.800 | 262k | – |
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0004 | – | 8.1 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0035 | – | 9.1 s |
| Code edit | 6,000 / 1,500 | $0.0029 | – | 11.9 s |
| Agentic coding session | 60,000 / 4,000 | $0.019 | – | 19.7 s |
| Structured extraction | 2,000 / 200 | $0.0007 | – | 7.8 s |
See also
Data as of 9 Sept 2026. Compare these configurations.