Benchmark Almanac

tau2-bench Verified

Grok 4.1 Fast 82.7% leads since December 2025
Grok 4.1 Fast 82.7% leads since December 2025
Fading

Agent serves simulated customers in airline, retail and telecom under policies, with Amazon-fixed faulty tasks from τ²-bench. Pass rate.

Lab-created verified variant (Amazon AGI) of Sierra's τ²-bench. Fixes documented in FIXES.md.

Made by
Amazon AGI
Released
December 2025
Kind
Customer service
Level
Everyday
Scored
Checked automatically
Answers
Agent task
Tasks
All public
Strongest model tested
GPT-5.2 #50 on the capability index
Last in a launch post
December 2025, dropped before it saturated
Last model added
December 2025
Scores refreshed
November 12th, 2026
14
Entries on its own leaderboard, 11 models we track
3
Launch posts list it, from 1 lab
3
System cards report it, from 1 lab

Grok 4.1 Fast leads with 82.7% among models released by December 2025

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

Grok 4.1 Fast leads with 82.7%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.

The newest model with a score here came out in December 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

14 entries. Scroll sideways for the rest.

  1. 82.7 Grok 4.1 Fast
  2. 82.0 Claude Opus 4.5
  3. 80.3 GPT-5.2
  4. 79.9 GPT-5
  5. 79.4 Gemini 3 Pro
  6. 78.5 Nova 2 Pro
  7. 76.8 GPT-5.1
  8. 75.7 Nova 2 Omni Preview
  9. 73.3 Claude Sonnet 4.5
  10. 72.4 Nova 2 Lite
  11. 69.8 GPT-5 mini
  12. 56.4 Gemini Pro 2.5
  13. 53.4 Claude Haiku 4.5
  14. 41.6 Gemini Flash 2.5
  1. 1 Grok 4.1 Fast reasoning 82.7%
  2. 2 Claude Opus 4.5 82%
  3. 3 GPT-5.2 xhigh 80.3%
  4. 4 GPT-5 med 79.9%
  5. 5 Gemini 3 Pro 79.4%
  6. 6 Nova 2 Pro 78.5%
  7. 7 GPT-5.1 high 76.8%
  8. 8 Nova 2 Omni Preview 75.7%
  9. 9 Claude Sonnet 4.5 73.3%
  10. 10 Nova 2 Lite 72.4%
  11. 11 GPT-5 mini med 69.8%
  12. 12 Gemini Pro 2.5 56.4%
  13. 13 Claude Haiku 4.5 53.4%
  14. 14 Gemini Flash 2.5 41.6%

Source: Official tau2-bench Verified leaderboard

Sources: Official tau2-bench Verified leaderboard and labs’ model cards

#EntryScoreSettingDateSource
1 Grok 4.1 Fast 82.7% reasoning, tau2-bench-verified; gpt-5.1 user simulator Nov 2025 Official leaderboard
2 Claude Opus 4.5 82% tau2-bench-verified; gpt-5.1 user simulator Nov 2025 Official leaderboard
3 GPT-5.2 80.3% xhigh, tau2-bench-verified; gpt-5.1 user simulator Dec 2025 Official leaderboard
4 GPT-5 79.9% med, tau2-bench-verified; gpt-5.1 user simulator Aug 2025 Official leaderboard
5 Gemini 3 Pro 79.4% tau2-bench-verified; gpt-5.1 user simulator Nov 2025 Official leaderboard
6 Nova 2 Pro 78.5% tau2-bench-verified; gpt-5.1 user simulator Official leaderboard
7 GPT-5.1 76.8% high, tau2-bench-verified; gpt-5.1 user simulator Nov 2025 Official leaderboard
8 Nova 2 Omni Preview 75.7% tau2-bench-verified; gpt-5.1 user simulator Dec 2025 Official leaderboard
9 Claude Sonnet 4.5 73.3% tau2-bench-verified; gpt-5.1 user simulator Sep 2025 Official leaderboard
10 Nova 2 Lite 72.4% tau2-bench-verified; gpt-5.1 user simulator Dec 2025 Official leaderboard
11 GPT-5 mini 69.8% med, tau2-bench-verified; gpt-5.1 user simulator Aug 2025 Official leaderboard
12 Gemini Pro 2.5 56.4% tau2-bench-verified; gpt-5.1 user simulator Official leaderboard
13 Claude Haiku 4.5 53.4% tau2-bench-verified; gpt-5.1 user simulator Oct 2025 Official leaderboard
14 Gemini Flash 2.5 41.6% tau2-bench-verified; gpt-5.1 user simulator Official leaderboard
#ModelScoreSettingDateSource
1 Grok 4.1 Fast 82.7% reasoning, tau2-bench-verified; gpt-5.1 user simulator Nov 2025 Official leaderboard
2 Claude Opus 4.5 82% tau2-bench-verified; gpt-5.1 user simulator Nov 2025 Official leaderboard
3 GPT-5.2 80.3% xhigh, tau2-bench-verified; gpt-5.1 user simulator Dec 2025 Official leaderboard
4 GPT-5 79.9% med, tau2-bench-verified; gpt-5.1 user simulator Aug 2025 Official leaderboard
5 Gemini 3 Pro 79.4% tau2-bench-verified; gpt-5.1 user simulator Nov 2025 Official leaderboard
6 GPT-5.1 76.8% high, tau2-bench-verified; gpt-5.1 user simulator Nov 2025 Official leaderboard
7 Nova 2 Omni Preview 75.7% tau2-bench-verified; gpt-5.1 user simulator Dec 2025 Official leaderboard
8 Claude Sonnet 4.5 73.3% tau2-bench-verified; gpt-5.1 user simulator Sep 2025 Official leaderboard
9 Nova 2 Lite 72.4% tau2-bench-verified; gpt-5.1 user simulator Dec 2025 Official leaderboard
10 Nova 2.0 Pro Preview 71.5% Dec 2025 Lab model card
11 GPT-5 mini 69.8% med, tau2-bench-verified; gpt-5.1 user simulator Aug 2025 Official leaderboard
12 Claude Haiku 4.5 53.4% tau2-bench-verified; gpt-5.1 user simulator Oct 2025 Official leaderboard

Labs report scores for 3 models in their cards. Switch the scores above to rank them together.

Number one over time Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

xAI Amazon
Own leaderboard Not tested since Dec 2025
Grok 4.1 Fast 82.7%
Lab claims Not tested since Dec 2025
Nova 2 Omni Preview 73.5%

Sources: Official tau2-bench Verified leaderboard and labs’ model cards

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Grok 4.1 Fast GPT-5 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Grok 4.1 Fast

Sources: Official tau2-bench Verified leaderboard and labs’ model cards

Labs headline it less often now, last in December 2025

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Amazon
DateLaunch postScoreAs the post writes it
Dec 2, 2025 Nova 2 Omni Preview 73.5 Mean of τ2-bench Retail-Verified 78.3 / Airline-Verified 68.8
Dec 2, 2025 Nova 2 Lite 70.7 Mean of τ2-bench Retail-Verified 76.5 / Airline-Verified 64.8 (avg@3)
Dec 2, 2025 Nova 2.0 Pro Preview 71.5 Mean of τ2-bench Retail-Verified 77.7 / Airline-Verified 65.2 (avg@3)

3 system cards report it, from 1 lab

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Dec 2, 2025 Nova 2 Omni Preview technical report tau2-bench Retail-Verified 78.3 and Airline-Verified 68.8, no overall (Table 5) Capabilities
Dec 2, 2025 Nova 2 Lite technical report tau2-bench Retail-Verified 76.5 and Airline-Verified 64.8, avg@3, no overall (Table 2) Capabilities
Dec 2, 2025 Nova 2.0 Pro Preview technical report tau2-bench Retail-Verified 77.7 and Airline-Verified 65.2, avg@3, no overall (Table 2) Capabilities

The lab's number comes out ahead in 0 of 2 pairs that differ

Each score a lab reported for its own model, beside the same model's best score on the leaderboard. Gaps mostly come from different harnesses, effort levels and task subsets. Every benchmark.

Model Lab says Board says Gap
Nova 2 Omni Preview 73.5% Mean of τ2-bench Retail-Verified 78.3 / Airline-Verified 68.8 75.7% Leaderboard, tau2-bench-verified; gpt-5.1 user simulator −2.2 pts
Nova 2 Lite 70.7% Mean of τ2-bench Retail-Verified 76.5 / Airline-Verified 64.8 (avg@3) 72.4% Leaderboard, tau2-bench-verified; gpt-5.1 user simulator −1.7 pts

More in Tool use

All 144 in Tool use
  • AutomationBench In use Agent completes end-to-end business workflows inside a simulated company exposing dozens of REST APIs. Graded on resulting state. Gemini 4 Carbon 53.9%
  • MCP Atlas In use Multi-step workflows over real MCP servers. Scores tool-use competency on realistic tasks. Gemini 4 Argon 90.7%
  • BFCL V4 In use Models call functions and tools, including multi-turn agentic tasks with web search and memory. Checked against expected calls and outcomes. Qwen3.7-Max 75%
  • τ³-bench Banking In use Agents act as banking customer-service reps over many turns with a simulated user, following policy. Pass^1 on the final database state. Qwen 3.8 Max 55.1%
  • Toolathlon-Verified In use 108 long-horizon tasks across 32 apps through about 600 MCP tools. Execution-based checkers inspect final environment state. GLM-5.3-Flash 78.4%
  • DeepSearchQA In use Questions that need multi-step search and synthesis. Final answers graded against references. Kimi K3 95%