BlogLLM Benchmark Scores as Published, Side by Side
τ-bench: Results as Published
Every result the catalog holds for τ-bench: 9 models from 3 developers, read from 4 pages. Each number is as printed by the company named beside it; nothing is ranked or rescored. Models are listed by developer, newest first.
Area Tool Use and WorkflowsMaintainer Sierra (Shunyu Yao and co-authors)Introduced 2024Data version 2026103001, as of 2026-10-30
tau-bench tests an agent that talks to a simulated user while using domain tools under policy rules, in an airline and a retail domain. A run passes when the final database state matches the annotated target; the score is the pass rate, with pass^k for repeated trials.
τ-bench Airline
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| o3OpenAI, released 2025-04-16 | 52% | Anthropic, Claude Opus 4 and Claude Sonnet 4checked 2026-10-15; printed on 1 more page | - |
| GPT-4.1OpenAI, released 2025-04-14 | 49.4% | Anthropic, Claude Opus 4 and Claude Sonnet 4checked 2026-10-15 | - |
| o1OpenAI, released 2024-12-17 | 54.2% | Anthropic, Claude 3.7 Sonnetchecked 2026-10-15 | - |
| Claude Opus 4.1Anthropic, released 2025-08-05 | 56% | Anthropic, Claude Opus 4.1checked 2026-10-15 | Effort: extended thinking, up to 64K |
| Claude Opus 4Anthropic, released 2025-05-22 | 59.6% | Anthropic, Claude Opus 4 and Claude Sonnet 4checked 2026-10-15; printed on 1 more page | Effort: extended thinking, up to 64K |
| Claude Sonnet 4Anthropic, released 2025-05-22 | 60% | Anthropic, Claude Opus 4 and Claude Sonnet 4checked 2026-10-15; printed on 1 more page | Effort: extended thinking, up to 64K |
| Claude 3.7 SonnetAnthropic, released 2025-02-24 | 58.4% | Anthropic, Claude 3.7 Sonnetchecked 2026-10-15; printed on 1 more page | Effort: none |
| Claude 3.5 SonnetAnthropic, released 2024-10-22 | 48.8% | Anthropic, Claude 3.7 Sonnetchecked 2026-10-15 | - |
| DeepSeek-R1-0528DeepSeek, released 2025-05-28 | 53.5% | DeepSeek, DeepSeek-R1-0528 model cardchecked 2026-10-15 | Harness: GPT-4.1 as the simulated user (card note) |
τ-bench Retail
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| o3OpenAI, released 2025-04-16 | 70.4% | Anthropic, Claude Opus 4 and Claude Sonnet 4checked 2026-10-15; printed on 1 more page | - |
| GPT-4.1OpenAI, released 2025-04-14 | 68% | Anthropic, Claude Opus 4 and Claude Sonnet 4checked 2026-10-15 | - |
| o1OpenAI, released 2024-12-17 | 73.5% | Anthropic, Claude 3.7 Sonnetchecked 2026-10-15 | - |
| Claude Opus 4.1Anthropic, released 2025-08-05 | 82.4% | Anthropic, Claude Opus 4.1checked 2026-10-15 | Effort: extended thinking, up to 64K |
| Claude Opus 4Anthropic, released 2025-05-22 | 81.4% | Anthropic, Claude Opus 4 and Claude Sonnet 4checked 2026-10-15; printed on 1 more page | Effort: extended thinking, up to 64K |
| Claude Sonnet 4Anthropic, released 2025-05-22 | 80.5% | Anthropic, Claude Opus 4 and Claude Sonnet 4checked 2026-10-15; printed on 1 more page | Effort: extended thinking, up to 64K |
| Claude 3.7 SonnetAnthropic, released 2025-02-24 | 81.2% | Anthropic, Claude 3.7 Sonnetchecked 2026-10-15; printed on 1 more page | Effort: none |
| Claude 3.5 SonnetAnthropic, released 2024-10-22 | 71.5% | Anthropic, Claude 3.7 Sonnetchecked 2026-10-15 | - |
| DeepSeek-R1-0528DeepSeek, released 2025-05-28 | 63.9% | DeepSeek, DeepSeek-R1-0528 model cardchecked 2026-10-15 | Harness: GPT-4.1 as the simulated user (card note) |
Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.