Trinity Large Thinking API Benchmarks, Pricing & Provider Data
Compare Trinity Large Thinking with another model
Choose a model to open its comparison page.
The page also shows measured API speed and first-token latency.
Trinity Large Thinking is a powerful open source reasoning model from the team at Arcee AI. It shows strong performance in PinchBench, agentic workloads, and reasoning tasks. Launch video: https://you...
Category Performance
Observed-capability estimates with uncertainty reported separately from independent benchmark families.
- Coverage
- 5 / 8
- Methodology
- V3.0
#1Knowledge56.4Provisional1/4 Measured dimensions
#2Coding49.7Provisional1/4 Measured dimensions
#3Instruction following48.2Provisional1/4 Measured dimensions
#4Reasoning44.4Estimated2/4 Measured dimensions
#5Agents43.9Estimated3/4 Measured dimensions
Specifications
Input and output token limits for this model, plus how it ranks on long-context understanding.
Features
Technical Details
- Input
- Output
- Released
- Apr 2026
- Tokenizer
- Other
- Architecture
- text->text
- Moderated
- No
- Supported parameters
- frequency_penaltyinclude_reasoninglogit_biaslogprobsmax_tokenspresence_penaltyreasoningrepetition_penaltyresponse_formatseedstopstructured_outputstemperaturetool_choicetoolstop_ktop_logprobstop_p
Rankings
Excels at
It's decent at
Falls behind in
Detailed scores
Updated: Aug 26, 2026Speed & latency
undefined metric} other undefined metrics}}
Output speed179.4 tok/s#13 / 75Time to first token0.67 s#14 / 75
Speed & latency
undefined metric} other undefined metrics}}
Pricing
undefined metric} other undefined metrics}}
Input price$0.235/M#42 / 179Output price$0.875/M#40 / 179
Pricing
undefined metric} other undefined metrics}}
Agents
V3.0undefined metric} other undefined metrics}} · Estimated
Score43.980% interval34.6–53.23/4 Measured dimensionsΤ²-bench results90.1#34 / 82Gert Labs32.5#44 / 49
Agents
V3.0undefined metric} other undefined metrics}} · Estimated
Coding
V3.0undefined metric} other undefined metrics}} · Provisional
Score49.780% interval33.7–65.71/4 Measured dimensionsSciCode36.1%#118 / 204SWE-bench Verified*63.2#5 / 5
Coding
V3.0undefined metric} other undefined metrics}} · Provisional
Reasoning
V3.0undefined metric} other undefined metrics}} · Estimated
Score44.480% interval33.6–55.22/4 Measured dimensionsGPQA75.2%#113 / 211HLE15.8%#93 / 208
Reasoning
V3.0undefined metric} other undefined metrics}} · Estimated
Knowledge
V3.0undefined metric} other undefined metrics}} · Provisional
Score56.480% interval40.4–72.41/4 Measured dimensionsGPQA-D76.3#31 / 31MMLU-Pro (Arcee)83.4#4 / 5
Knowledge
V3.0undefined metric} other undefined metrics}} · Provisional
Math
V3.0undefined metric} other undefined metrics}} · No data
80% interval30.8–69.20/4 Measured dimensionsAIME25 (Arcee)96.3#2 / 5
Math
V3.0undefined metric} other undefined metrics}} · No data
Multilingual
V3.0undefined metric} other undefined metrics}} · No data
No data80% interval30.8–69.20/4 Measured dimensions
Multilingual
V3.0undefined metric} other undefined metrics}} · No data
Multimodal
V3.0undefined metric} other undefined metrics}} · No data
80% interval30.8–69.20/4 Measured dimensionsDesign Arena Website1148.0#59 / 75
Multimodal
V3.0undefined metric} other undefined metrics}} · No data
Instruction following
V3.0undefined metric} other undefined metrics}} · Provisional
Score48.280% interval32.2–64.21/4 Measured dimensionsAA-IFBench56.3#53 / 84
Instruction following
V3.0undefined metric} other undefined metrics}} · Provisional
OpenRouter endpoints
2 endpointsThird-party OpenRouter endpoint data, shown separately from LMSpeed measurements. Some 30-minute live performance fields only appear after syncing with an OpenRouter API key.
| Provider endpoint | Input | Output | 1d uptime | 30m latency | 30m throughput | Context / output |
|---|---|---|---|---|---|---|
Parasail parasail/fp4 | $0.220/M | $0.850/M | 100.0% | — | — | undefined tokens / undefined tokens |
Arcee AI arcee-ai | $0.250/M | $0.800/M | 99.9% | — | — | undefined tokens / undefined tokens |
Alternatives & Similar Models
Claude Haiku 4.5
claude-haiku-4-5
Anthropic Claude Haiku 4.5 delivers fast, low-cost responses while retaining solid instruction following for chat, classification, and lightweight coding.
Kimi K2.5
kimi-k2-5
Moonshot Kimi K2.5 is an open-weight multimodal agent model with native vision and text input, strong coding performance, and a 256K context window.
MiniMax M2.5
minimax-m2-5
MiniMax M2.5 is MiniMax's flagship text model for coding and agents, with SOTA-level programming and agentic performance, improved token efficiency, and fast high-TPS API deployment.
Claude Opus 4.6
claude-opus-4-6
Anthropic Claude Opus 4.6 is the most capable Claude Opus tier, optimized for complex analysis, long-horizon coding, and high-stakes enterprise reasoning workloads.
DeepSeek V4 Flash
deepseek-v4-flash
DeepSeek V4 Flash is a fast, cost-efficient language model in the DeepSeek V4 family, optimized for low-latency chat, coding assistance, and high-throughput API workloads while retaining strong reasoning quality.
Claude Opus 4.7
claude-opus-4-7
Anthropic Claude Opus 4.7 targets frontier-level analysis, complex coding, and autonomous workflows that require deep multi-step reasoning.