reasoning benchmark
t2-bench Leaderboard
t2-bench is a reasoning benchmark for evaluating agentic tool use in complex, multi-step tasks.
Updated Sep 5, 2026
Models23
Model coverage23
MetricScore
EvidenceB
t2-bench Ranking
Higher score ranks better on this benchmark.
Sort by
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelGO | Score99.30% | Percentile100.00% | Participants23 | EvidenceC | Evaluated |
| Rank02 | ModelGO | Score90.20% | Percentile95.45% | Participants23 | EvidenceC | Evaluated |
| Rank03 | ModelZA | Score89.70% | Percentile90.91% | Participants23 | EvidenceC | Evaluated |
| Rank04 | ModelAC | Score86.70% | Percentile86.36% | Participants23 | EvidenceC | Evaluated |
| Rank05 | ModelGO | Score86.40% | Percentile81.82% | Participants23 | EvidenceC | Evaluated |
| Rank06 | ModelGO | Score85.50% | Percentile77.27% | Participants23 | EvidenceC | Evaluated |
| Rank07 | ModelGO | Score85.40% | Percentile72.73% | Participants23 | EvidenceC | Evaluated |
| Rank08 | ModelAC | Score81.20% | Percentile68.18% | Participants23 | EvidenceC | Evaluated |
| Rank09 | ModelDE | Score80.30% | Percentile63.64% | Participants23 | EvidenceC | Evaluated |
| Rank10 | ModelDE | Score80.30% | Percentile59.09% | Participants23 | EvidenceC | Evaluated |
| Rank11 | ModelDE | Score80.20% | Percentile54.55% | Participants23 | EvidenceC | Evaluated |
| Rank12 | ModelAC | Score79.90% | Percentile50.00% | Participants23 | EvidenceC | Evaluated |
| Rank13 | ModelAC | Score79.50% | Percentile45.45% | Participants23 | EvidenceC | Evaluated |
| Rank14 | ModelAC | Score79.10% | Percentile40.91% | Participants23 | EvidenceC | Evaluated |
| Rank15 | ModelAC | Score79.00% | Percentile36.36% | Participants23 | EvidenceC | Evaluated |
| Rank16 | ModelAC | Score74.80% | Percentile31.82% | Participants23 | EvidenceC | Evaluated |
| Rank17 | ModelLA | Score73.20% | Percentile27.27% | Participants23 | EvidenceC | Evaluated |
| Rank18 | ModelOP | Score63.90% | Percentile22.73% | Participants23 | EvidenceC | Evaluated |
| Rank19 | ModelGO | Score57.50% | Percentile18.18% | Participants23 | EvidenceC | Evaluated |
| Rank20 | ModelGO | Score56.20% | Percentile13.64% | Participants23 | EvidenceC | Evaluated |
| Rank21 | ModelAC | Score48.80% | Percentile9.09% | Participants23 | EvidenceC | Evaluated |
| Rank22 | ModelGO | Score29.40% | Percentile4.55% | Participants23 | EvidenceC | Evaluated |
| Rank23 | ModelAC | Score11.60% | Percentile0.00% | Participants23 | EvidenceC | Evaluated |
t2-bench Highlights
The leading models and scores on this benchmark.
t2-bench Score Distribution
A closer view of the leading scores on this benchmark.
The Top AI Models for t2-bench
The first five results on this benchmark, with official price and output speed added where the model identity can be matched.
What is t2-bench?
What t2-bench measures and how its scores work.
t2-bench is a benchmark that evaluates agentic tool use capabilities.
It measures how well models select, sequence, and utilize tools, as well as their autonomous planning and execution in multi-step scenarios, using the Score metric with a ratio unit.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- t2-bench
- Modality
- text
- Primary category
- reasoning
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- overall
LLM Stats. Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about t2-bench.
Which model scores highest on t2-bench?
Gemini 3.1 Pro is currently ranked first with 99.30%.
What are the top three models on t2-bench?
The current leaders are Gemini 3.1 Pro (99.30%), Gemini 3 Flash (90.20%), and GLM-5 (89.70%).
Which t2-bench model has the lowest official input price?
Qwen3.5-35B-A3B has the lowest matched official input price at $0.25 input / $2.0 output per 1M tokens.
Which models are fastest among t2-bench results?
The fastest matched records are Gemini 3 Flash (124.32 tok/s via Google), GPT OSS 120B High (100.00 tok/s via OpenAI), and DeepSeek-V3.2 (97.00 tok/s via DeepInfra).
Does the highest score result prove overall model quality?
No. This benchmark measures one defined capability or task. The overall LLMBoard score uses a separate aggregation across eligible benchmark evidence.