ToolMaze
Agents complete tool-use tasks while the runtime introduces explicit or hidden tool failures. A judge checks completion across four task structures and five perturbation settings. Metrics report task success, recovery rate, and recovery cost.
Recovery rate includes only runs that encountered a perturbed call. Recovery cost compares actual calls with an oracle recovery path. The runtime includes 273 tool plugins.
- Released
- June 2026
- Kind
- Function calling
- Level
- University
- Scored
- Graded by a model
- Answers
- Agent task
- Tasks
- All public
- Measures
- Capability
- Language
- English
- Interaction
- Agentic
- Environment
- APIs and tools
- Items from
- Generated
- Refresh
- One fixed set
- Run by
- Anyone
- In views
- Agents
- Strongest model tested
- GPT-5.5 #21 on the capability index
- Last model added
- April 2026
- Scores refreshed
- November 19th, 2026
In the authors’ words
Existing benchmarks evaluate Tool-Integrated Reasoning (TIR) in LLMs on idealized ”happy paths”, largely overlooking real-world tool failures. We introduce ToolMaze, a benchmark for dynamic path discovery and error recovery in TIR agents. To separate systematic replanning from blind trial-and-error, ToolMaze adopts a two-dimensional design: DAG-based topological complexity and a 2 × 2 taxonomy of tool perturbations (explicit/implicit, transient/permanent). Evaluations show that perturbations degrade performance across nearly all models, with the sharpest drops under implicit semantic failures. Driven by systemic over-trust in corrupted outputs, Perturbation Recovery Rate (PRR) plummets by around 37% in these scenarios, while complex topologies trap agents in futile trial-and-error loops. Crucially, agentic fault-tolerance improves with model scale 3.66× slower than basic task execution, highlighting dynamic replanning as a distinct bottleneck unaddressed by model scaling or prompting. Data and code are available at https://github.com/Zhudongsheng75/ToolMaze.
Gemini 3.1 Pro leads with 53% among models released by April 2026
Best score per model from the benchmark's paper, read on November 19th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.
The newest model with a score here came out in April 2026, 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 3.1 Pro w/ hint 53%
- 2 GPT-5.5 w/ hint 50%
- 3 Claude Sonnet 4.6 w/ hint 46.1%
- 4
DeepSeek-V4-Pro w/ hint 45.1%
- 5 GLM-5.1 w/ hint 38.1%
- 6
Qwen3.5 397B-A17B w/ hint 35.6%
- 7
Qwen3.6 27B w/ hint 34.1%
- 8
Qwen3.5-35B-A3B w/ hint 32.6%
- 9
MiniMax-M2.7 w/ hint 27.5%
Source: The ToolMaze paper
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 53% | w/ hint | Feb 2026 | Official leaderboard |
| 2 | GPT-5.5 | 50% | w/ hint | Apr 2026 | Official leaderboard |
| 3 | Claude Sonnet 4.6 | 46.1% | w/ hint | Feb 2026 | Official leaderboard |
| 4 | | 45.1% | w/ hint | Apr 2026 | Official leaderboard |
| 5 | GLM-5.1 | 38.1% | w/ hint | Apr 2026 | Official leaderboard |
| 6 | | 35.6% | w/ hint | Feb 2026 | Official leaderboard |
| 7 | | 34.1% | w/ hint | Apr 2026 | Official leaderboard |
| 8 | | 32.6% | w/ hint | Feb 2026 | Official leaderboard |
| 9 | | 27.5% | w/ hint | Mar 2026 | Official leaderboard |
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 ToolMaze paper
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| w/ hint | 53% | ||||
| GPT-5.5 | w/ hint | 50% | |||
| Claude Sonnet 4.6 | w/ hint | 46.1% | |||
| w/ hint | 45.1% | ||||
| GLM-5.1 | w/ hint | 38.1% | |||
| w/o hint | 36.3% | ||||
| GPT-5.5 | w/o hint | 35.9% | |||
| w/ hint | 35.6% | ||||
| w/ hint | 34.1% | ||||
| w/ hint | 32.6% | ||||
| GLM-5.1 | w/o hint | 32.3% | |||
| w/o hint | 32.2% | ||||
| w/o hint | 31.6% | ||||
| w/o hint | 31.6% | ||||
| w/o hint | 29.6% | ||||
| w/ hint | 27.5% | ||||
| Claude Sonnet 4.6 | w/o hint | 27.4% | |||
| w/o hint | 25% |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The ToolMaze paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official ToolMaze 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.
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
9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
- Pass Independent runs Scores from the benchmark’s paper (9 entries).
- Pass Models scored 9 models we track.
- Warning Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.5 (#21).
- Pass Headroom The best is 53%, under 85%, so it still separates models.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 18 rows for models we track at the last read.
- Pass Last read November 19th, 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.