τc-Bench
Version Contextual
Agents use tools while conversing with simulated customers in airline, retail, and telecom tasks. The evaluator compares the final environment state with the expected state. Generated scenarios add unusual tool combinations, ambiguous requests, and distracting database records.
This extension changes the tasks from τ²-Bench rather than only fixing them. The TASTE synthesis method is unrelated to Anthropic’s TASTE benchmark.
- Made by
- Technion and IBM Research
- Released
- May 2026
- Kind
- Customer service
- Level
- Everyday
- Scored
- Checked automatically
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Agentic
- Environment
- APIs and tools
- Items from
- Generated, Another benchmark
- Refresh
- New versions replace it
- Run by
- Anyone
- Based on
- tau2-bench
- In views
- Agents
- Strongest model tested
- GPT-5.2 #51 on the capability index
- Last model added
- December 2025
- Scores refreshed
- November 20th, 2026
In the authors’ words
As agent capabilities advance, existing benchmarks, such as τ²-Bench, are becoming increasingly saturated. Yet constructing new benchmark tasks remains complex, costly, and labor-intensive. Moreover, the standard approach, in which scenarios are first written in natural language and then mapped to tool sequences, captures only a narrow subset of the tool-use patterns agents exercise. In this paper, we address these problems by reversing the task construction process. We propose TASTE: Task Synthesis from Tool Sequence Evolution, an automatic method that generates challenging tasks with broader tool-use coverage. TASTE utilizes an Adaptive Contrastive n-gram model trained on LLM-judged validity signals. This enables sampling valid tool sequences that cover a vast range of tool combinations. TASTE then selects representative sequences from the pool via clustering, instantiates them into complete benchmark tasks, and refines them through iterative difficulty evolution. Using TASTE, we construct τc-Bench, a challenging extension of the three domains of τ²-Bench. We evaluate 11 agent/user LLM pairs and find that models nearly saturating τ²-Bench suffer severe performance drops on our tasks (e.g., Gemini-3-Flash falls from 0.82 - 0.94 to 0.28 - 0.61). Beyond increasing difficulty, our generated tasks more than double the number of unique tool combinations agents must execute. Our results suggest high scores on existing benchmarks often reflect saturation rather than robust task-solving ability. By automating the generation of difficult, high-coverage benchmarks, TASTE enables continuous, scalable evaluation of future agents.
Gemini 3 Flash leads with 87.7% among models released by December 2025 Gemini 3 Flash leads with 87.7% among models released by December 2025
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper reports no overall score, so this is the plain mean of Airline, Retail and Telecom pass^1 scores. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
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:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1
Gemini 3 Flash user simulator: GPT-5.2 87.7%
- 2 GPT-5.2 user simulator: Gemini-3-flash 68%
- 3 deepseek-3.1 user simulator: GPT-5.2 65%
- 4
Gemini 2.5 Flash Sep 2025, user simulator: Gemini-3-flash 57.3%
- 5 Qwen-32B user simulator: Gemini-3-flash 46%
Source: The τc-Bench paper
Source: The τc-Bench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 87.7% | user simulator: GPT-5.2 | Dec 2025 | Official leaderboard |
| 2 | GPT-5.2 | 68% | user simulator: Gemini-3-flash | Dec 2025 | Official leaderboard |
| 3 | deepseek-3.1 | 65% | user simulator: GPT-5.2 | Official leaderboard | |
| 4 | | 57.3% | user simulator: Gemini-3-flash | Sep 2025 | Official leaderboard |
| 5 | Qwen-32B | 46% | user simulator: Gemini-3-flash | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 87.7% | user simulator: GPT-5.2 | Dec 2025 | Official leaderboard |
| 2 | GPT-5.2 | 68% | user simulator: Gemini-3-flash | Dec 2025 | Official leaderboard |
| 3 | | 57.3% | user simulator: Gemini-3-flash | Sep 2025 | Official leaderboard |
Rankings by part
We read 3 parts from the benchmark’s paper: Airline pass^1, Retail pass^1 and Telecom pass^1. Each ranks models on one part of the test. Gemini 3 Flash leads 2 of them, and GPT-5.2 leads the other one. Arrows show how far a model moved from its overall rank among the models we track.
Airline pass^1
3 models we track have a score here.
- 1
Gemini 3 Flash 82%
- 2
Gemini 2.5 Flash Sep 2025 ↑1 66%
- 3 GPT-5.2 ↓1 58%
Retail pass^1
3 models we track have a score here.
- 1 GPT-5.2 ↑1 92%
- 2
Gemini 3 Flash ↓1 88%
- 3
Gemini 2.5 Flash Sep 2025 79%
Telecom pass^1
3 models we track have a score here.
- 1
Gemini 3 Flash 94%
- 2 GPT-5.2 62%
- 3
Gemini 2.5 Flash Sep 2025 46%
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: The τc-Bench paper
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| user simulator: GPT-5.2 | 87.7% | ||||
| user simulator: Gemini-3-flash | 83.3% | ||||
| GPT-5.2 | user simulator: Gemini-3-flash | 68% | |||
| GPT-5.2 | user simulator: GPT-5.2 | 63% | |||
| user simulator: Gemini-3-flash | 57.3% | ||||
| user simulator: GPT-5.2 | 57.3% |
Scores at each effort level
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: The τc-Bench paper
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| user simulator: GPT-5.2 | 87.7% | ||||
| user simulator: Gemini-3-flash | 83.3% | ||||
| GPT-5.2 | user simulator: Gemini-3-flash | 68% | |||
| GPT-5.2 | user simulator: GPT-5.2 | 63% | |||
| user simulator: Gemini-3-flash | 57.3% | ||||
| user simulator: GPT-5.2 | 57.3% |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The τc-Bench paper
Gemini 3 Flash on the official leaderboard passed 85% in December 2025
Models released before it came out already scored that high. All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official τc-Bench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
Versions
We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.
Sources: Official leaderboards and labs’ model cards
More 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 Neon 56.3%
- 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%
Data audit
All 8 checks pass. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
- Pass Independent runs Scores from the benchmark’s paper (5 entries).
- Pass Models scored 3 models we track, plus 2 other entries on its leaderboard.
- Note Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. That’s expected for a test replaced by τ-bench 2.0.
- Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is GPT-5.2 (#51).
- Note Headroom Saturated. The best is 87.7%, past 85% since December 2025.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 6 rows for models we track and 2 other entries at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.