Thinkingbox-bench
Agents complete 507 business workflows through tools and simulated user conversations. Executable checks require correct backend updates without forbidden side effects.
Some tasks also grade final responses against rubrics. The paper reports repeated-attempt reliability, but the linked dataset repository returned 404 during this audit.
- Made by
- Microsoft and collaborators
- Released
- August 2026
- Kind
- Customer service
- Level
- University
- Scored
- Checked automatically
- Answers
- Agent task
- Measures
- Capability
- Language
- English
- Interaction
- Agentic
- Environment
- APIs and tools
- Items from
- Generated
- Refresh
- One fixed set
- Run by
- A host
- In views
- Agents
- Strongest model tested
- GPT-6 Astra #4 on the capability index
- Last model added
- September 2026
- Scores refreshed
- November 20th, 2026
In the authors’ words
Recent agent benchmarks increasingly ground evaluation in executable environments, from code repair to web navigation, app APIs, and function calling. Yet completing consequential work beyond code requires more than producing a plausible response or valid tool call: agents must gather missing information over multiple turns, follow domain policies, coordinate dependent tools, and realize the correct persistent state transition without collateral effects. In this paper, we introduce Thinkingbox, a sandbox for tool-agent-user interaction that provides isolated MCP-compatible tool sessions, complete execution traces, and outcome evaluation over terminal backend state. Built on this sandbox, Thinkingbox-bench contains 507 policy-conditioned workflows across business scenarios, including retail, hospitality, auto insurance, neobank internal IT, and consulting IT/HR support. Each attempt is evaluated by task-specific executable checks that accept valid trajectories while rejecting wrong, missing, or extra effects; designated tasks additionally check required properties of the final response. Our experiments reveal that even the strongest proprietary and open-weight models show steep reliability drops: Claude Opus 5 falls from 66.50% pass@1 to 47.53% pass^20, and Kimi-K3 from 57.37% pass@1 to 17.60% pass^20. Moreover, many failed trials terminate cleanly after valid state-changing actions, so response- or tool-call-level signals poorly proxy end-to-end completion. Thinkingbox-bench reveals a large gap between occasionally finding a successful trajectory and reliably completing stateful business tasks. We release Thinkingbox, Thinkingbox-Bench, and Thinkingbox-integrated reinforcement learning framework.
Claude Opus 5 leads with 66.5% among models released by September 2026
Best score per model from the benchmark's paper, read on November 20th, 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 September 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
- Claude Opus 5.5
- and 13 more
- 66.5 Claude Opus 5
- 65.4 GPT-5.4
- 61.9 GPT-5.6 Sol
- 59.2 Claude Sonnet 4.6
- 58.3 GPT-6 Astra
- 57.4 Kimi K3
- 51.7
Qwen 3.8 27B
- 46.3 GPT-5.2
- 43.3
DeepSeek-V4-Pro
- 37.7 Kimi K2.6
- 33.2 GLM-5.1
- 32.9
Qwen3.6 27B
- 32.1 Claude Opus 4.6
- 19.3 o3-pro
- 14.4 Grok 4.3
- 5.65
Qwen3.5-9B
- 4.66
Mistral Large 3
- 0.19
MiniMax-M2.5
- 1 Claude Opus 5 507 tasks; 20 trials; medium 66.5%
- 2 GPT-5.4 507 tasks; 20 trials; medium 65.4%
- 3 GPT-5.6 Sol 507 tasks; 20 trials; medium 61.9%
- 4 Claude Sonnet 4.6 507 tasks; 20 trials; medium 59.2%
- 5 GPT-6 Astra 507 tasks; 20 trials; medium 58.3%
- 6 Kimi K3 507 tasks; 20 trials; medium 57.4%
- 7
Qwen 3.8 27B 507 tasks; 20 trials; medium 51.7%
- 8 GPT-5.2 507 tasks; 20 trials; medium 46.3%
- 9
DeepSeek-V4-Pro 507 tasks; 20 trials; medium 43.3%
- 10 Kimi K2.6 507 tasks; 20 trials; medium 37.7%
- 11 GLM-5.1 507 tasks; 20 trials; medium 33.2%
- 12
Qwen3.6 27B 507 tasks; 20 trials; medium 32.9%
- 13 Claude Opus 4.6 507 tasks; 20 trials; medium 32.1%
- 14 o3-pro 507 tasks; 20 trials; medium 19.3%
- 15 Grok 4.3 507 tasks; 20 trials; medium 14.4%
- 16
Qwen3.5-9B 507 tasks; 20 trials; medium 5.65%
- 17
Mistral Large 3 507 tasks; 20 trials; medium 4.66%
- 18
MiniMax-M2.5 507 tasks; 20 trials; medium 0.19%
Source: The Thinkingbox-bench paper
| # | Model | Score ± | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Opus 5 | 66.5% ±3.7 | 507 tasks; 20 trials; medium, Thinkingbox | Jul 2026 | Official leaderboard |
| 2 | GPT-5.4 | 65.4% ±3.1 | 507 tasks; 20 trials; medium, Thinkingbox | Mar 2026 | Official leaderboard |
| 3 | GPT-5.6 Sol | 61.9% ±3.3 | 507 tasks; 20 trials; medium, Thinkingbox | Jul 2026 | Official leaderboard |
| 4 | Claude Sonnet 4.6 | 59.2% 55.2 to 61.7 | 507 tasks; 20 trials; medium, Thinkingbox | Feb 2026 | Official leaderboard |
| 5 | GPT-6 Astra | 58.3% ±4 | 507 tasks; 20 trials; medium, Thinkingbox | Sep 2026 | Official leaderboard |
| 6 | Kimi K3 | 57.4% ±3.1 | 507 tasks; 20 trials; medium, Thinkingbox | Jul 2026 | Official leaderboard |
| 7 | | 51.7% ±3.1 | 507 tasks; 20 trials; medium, Thinkingbox | Aug 2026 | Official leaderboard |
| 8 | GPT-5.2 | 46.3% ±3.2 | 507 tasks; 20 trials; medium, Thinkingbox | Dec 2025 | Official leaderboard |
| 9 | | 43.3% ±3.1 | 507 tasks; 20 trials; medium, Thinkingbox | Apr 2026 | Official leaderboard |
| 10 | Kimi K2.6 | 37.7% ±2.8 | 507 tasks; 20 trials; medium, Thinkingbox | Apr 2026 | Official leaderboard |
| 11 | GLM-5.1 | 33.2% ±3 | 507 tasks; 20 trials; medium, Thinkingbox | Apr 2026 | Official leaderboard |
| 12 | | 32.9% ±2.8 | 507 tasks; 20 trials; medium, Thinkingbox | Apr 2026 | Official leaderboard |
| 13 | Claude Opus 4.6 | 32.1% ±3.5 | 507 tasks; 20 trials; medium, Thinkingbox | Feb 2026 | Official leaderboard |
| 14 | o3-pro | 19.3% ±2.3 | 507 tasks; 20 trials; medium, Thinkingbox | Jun 2025 | Official leaderboard |
| 15 | Grok 4.3 | 14.4% ±2.2 | 507 tasks; 20 trials; medium, Thinkingbox | Apr 2026 | Official leaderboard |
| 16 | | 5.65% ±1.2 | 507 tasks; 20 trials; medium, Thinkingbox | Feb 2026 | Official leaderboard |
| 17 | | 4.66% 3.7 to 5.8 | 507 tasks; 20 trials; medium, Thinkingbox | Dec 2025 | Official leaderboard |
| 18 | | 0.19% 0 to 0.5 | 507 tasks; 20 trials; medium, Thinkingbox | Feb 2026 | Official leaderboard |
Rankings by part
We read 5 parts from the benchmark’s paper, each ranking models on one part of the test, such as Retail, Auto and Booking. Claude Opus 5 leads 3 of them, and 2 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.
Retail
14 models we track have a score here.
- 1 Kimi K3 ↑5 82.2%
- 2 Claude Opus 5 ↓1 80.7%
- 3 GPT-5.4 ↓1 76.3%
- 4 Claude Sonnet 4.6 72.3%
- 5 GPT-6 Astra 71.7%
- 6 Claude Opus 4.6 ↑7 68.6%
- 7
DeepSeek-V4-Pro ↑2 68.2%
- 8 GPT-5.6 Sol ↓5 67.7%
- 9
Qwen 3.8 27B ↓2 64%
- 10 GLM-5.1 ↑1 58.7%
Auto
14 models we track have a score here.
- 1 Claude Opus 5 65.8%
- 2 GPT-5.6 Sol ↑1 65.3%
- 3 GPT-5.4 ↓1 62.6%
- 4 Claude Sonnet 4.6 54.4%
- 5 Kimi K3 ↑1 50.8%
- 6
Qwen 3.8 27B ↑1 47.9%
- 7 GPT-6 Astra ↓2 46.5%
- 8
DeepSeek-V4-Pro ↑1 29.6%
- 9 GLM-5.1 ↑2 25.7%
- 10 Claude Opus 4.6 ↑3 8.30%
Booking
14 models we track have a score here.
- 1 GPT-5.4 ↑1 68.1%
- 2 Kimi K3 ↑4 61.8%
- 3 GPT-5.6 Sol 60.3%
- 4 Claude Sonnet 4.6 58.9%
- 5 GPT-6 Astra 55.9%
- 6
Qwen 3.8 27B ↑1 53.4%
- 7 Claude Opus 5 ↓6 50%
- 8
DeepSeek-V4-Pro ↑1 43.1%
- 9 GLM-5.1 ↑2 35.4%
- 10 Claude Opus 4.6 ↑3 21.1%
Bank
14 models we track have a score here.
- 1 Claude Opus 5 70.6%
- 2 GPT-5.4 65.3%
- 3 GPT-6 Astra ↑2 60.9%
- 4 GPT-5.6 Sol ↓1 59.1%
- 5 Claude Sonnet 4.6 ↓1 56.4%
- 6
Qwen 3.8 27B ↑1 47.9%
- 7
DeepSeek-V4-Pro ↑2 44.9%
- 8 Kimi K3 ↓2 41.4%
- 9 Claude Opus 4.6 ↑4 35.7%
- 10 o3-pro ↑4 24.3%
Consulting
14 models we track have a score here.
- 1 Claude Opus 5 66.2%
- 2 GPT-5.6 Sol ↑1 57.5%
- 3 GPT-6 Astra ↑2 56.8%
- 4 GPT-5.4 ↓2 54.6%
- 5 Claude Sonnet 4.6 ↓1 54.3%
- 6 Kimi K3 51.6%
- 7
Qwen 3.8 27B 45.7%
- 8 GLM-5.1 ↑3 34.1%
- 9
DeepSeek-V4-Pro 31%
- 10 Claude Opus 4.6 ↑3 27.8%
The best score, 66.5%, costs $0.32 a task
Runs on the cost frontier score higher than any run that's cheaper. GPT-5.6 Sol at 507 tasks; 20 trials; medium comes within 4.6 points of the best for $0.08 a task, 4 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.
Source: The Thinkingbox-bench paper
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Claude Opus 5 | 507 tasks; 20 trials; medium | Thinkingbox | 66.5% | $0.32 | |
| GPT-5.4 | 507 tasks; 20 trials; medium | Thinkingbox | 65.4% | $0.09 | |
| GPT-5.6 Sol | 507 tasks; 20 trials; medium | Thinkingbox | 61.9% | $0.08 | |
| Claude Sonnet 4.6 | 507 tasks; 20 trials; medium | Thinkingbox | 59.2% | $0.16 | |
| GPT-6 Astra | 507 tasks; 20 trials; medium | Thinkingbox | 58.3% | $0.17 | |
| Kimi K3 | 507 tasks; 20 trials; medium | Thinkingbox | 57.4% | $0.14 | |
| 507 tasks; 20 trials; medium | Thinkingbox | 51.7% | $0.09 | ||
| GPT-5.2 | 507 tasks; 20 trials; medium | Thinkingbox | 46.3% | $0.09 | |
| 507 tasks; 20 trials; medium | Thinkingbox | 43.3% | $0.22 | ||
| Kimi K2.6 | 507 tasks; 20 trials; medium | Thinkingbox | 37.7% | $0.05 | |
| GLM-5.1 | 507 tasks; 20 trials; medium | Thinkingbox | 33.2% | $0.11 | |
| 507 tasks; 20 trials; medium | Thinkingbox | 32.9% | $0.11 | ||
| Claude Opus 4.6 | 507 tasks; 20 trials; medium | Thinkingbox | 32.1% | $0.25 | |
| o3-pro | 507 tasks; 20 trials; medium | Thinkingbox | 19.3% | $3.63 | |
| Grok 4.3 | 507 tasks; 20 trials; medium | Thinkingbox | 14.4% | $0.06 | |
| 507 tasks; 20 trials; medium | Thinkingbox | 5.65% | $0.06 | ||
| 507 tasks; 20 trials; medium | Thinkingbox | 4.66% | $0.06 | ||
| 507 tasks; 20 trials; medium | Thinkingbox | 0.19% | $0.006 |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The Thinkingbox-bench 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 Thinkingbox-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.
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 Task access check. We don’t know whether the tasks are public. Every benchmark
Catalog record 1 warning 1 pass
- Warning Task access We don’t know whether the tasks are public.
- Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
- Pass Independent runs Scores from the benchmark’s paper (18 entries).
- Pass Models scored 18 models we track.
- Pass Newest score The newest is from September 2026, for GPT-6 Astra.
- Note Frontier models Only 1 of the top 10 has a score. The strongest tested is GPT-6 Astra (#4).
- Pass Headroom The best is 66.5%, 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 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.