Skip to main content

Benchmark profile

PinchBench

An OpenClaw agent benchmark from Kilo that measures successful task completion across standardized real-world agent workflows.

The public PinchBench snapshot ranks anthropic/claude-opus-4.8-fast first at 93.5%, ahead of Qwen3.7 Max (92.5%) and Claude Opus 4.8 (90.5%) among 58 tested models. We mirror the table as display-only evidence; it does not affect overall rankings.

How BenchLM shows PinchBench

BenchLM mirrors the public PinchBench average-success-rate view using the official snapshot updated on 07/10/2026, 12:22 AM: 58 models and 614 runs. PinchBench grades runs with automated checks plus an LLM judge.

This benchmark is display only on BenchLM. It is excluded from BenchLM overall rankings, category rankings, and weighted scoring. The table below uses average scores only, matching the public PinchBench average view rather than the best-run view.

58 models614 runsAverage scores onlyOfficial runsDisplay only

Average success rate on PinchBench — 07/10/2026, 12:22 AM

BenchLM mirrors the published average success rate view for PinchBench. anthropic/claude-opus-4.8-fast leads the public snapshot at 93.5% , followed by Qwen3.7 Max (92.5%) and Claude Opus 4.8 (90.5%). BenchLM does not use these results to rank models overall.

58 modelsAgenticCurrentDisplay onlyUpdated 07/10/2026, 12:22 AM

Average success rate table (58 models)

Score
2
Qwen3.7 MaxAlibaba · Closed
92.5%
3
Claude Opus 4.8Anthropic · Closed
90.5%
5
MiMo-V2.5Xiaomi · Closed
89.7%
6
Grok Build 0.1xAI · Closed
88.9%
7
GPT-5.6 LunaOpenAI · Closed
88.7%
8
88.1%
9
MiMo-V2.5-ProXiaomi · Closed
87.5%
10
GLM-5.2Z.AI · Open weight
87.0%
11
GPT-5.6 SolOpenAI · Closed
84.2%
12
82.6%
13
DeepSeek V4 FlashDeepSeek · Open weight
81.7%
14
Gemini 3.1 ProGoogle · Closed
81.0%
15
Gemini 3.1 Flash-LiteGoogle · Closed
80.5%
16
Grok 4.20xAI · Closed
80.3%
17
Step 3.5 FlashStepFun · Open weight
79.4%
18
GPT-5.4 miniOpenAI · Closed
79.2%
19
Kimi K2.7 CodeMoonshot AI · Open weight
76.1%
20
Claude Opus 4.7Anthropic · Closed
76.0%
21
GPT-5.6 TerraOpenAI · Closed
75.9%
22
GPT-5.4OpenAI · Closed
75.7%
23
GPT-5.5OpenAI · Closed
75.5%
24
Grok 4.5xAI · Closed
75.2%
25
Seed-2.0-LiteByteDance · Closed
75.0%
26
Gemini 3.5 FlashGoogle · Closed
74.2%
27
Grok 4.3xAI · Closed
73.7%
28
Qwen3.6 PlusAlibaba · Closed
72.5%
29
Gemini 3 FlashGoogle · Closed
72.1%
30
GLM-5-TurboZ.AI · Closed
71.8%
31
Claude Opus 4.6Anthropic · Closed
69.9%
32
69.5%
34
GPT-5.4 nanoOpenAI · Closed
69.0%
35
Claude Haiku 4.5Anthropic · Closed
67.7%
36
GLM-5V-TurboZ.AI · Closed
67.6%
37
MiniMax M2.7MiniMax · Open weight
66.8%
38
Trinity-Large-PreviewArcee AI · Open weight
65.7%
40
Claude Sonnet 4.6Anthropic · Closed
62.7%
41
61.1%
42
DeepSeek V4 ProDeepSeek · Open weight
61.1%
43
GLM-5.1Z.AI · Open weight
59.9%
45
Claude Fable 5Anthropic · Closed
54.8%
46
Kimi K2.5Moonshot AI · Open weight
54.6%
48
Trinity-Large-PreviewArcee AI · Open weight
53.0%
51
GPT-OSS 120BOpenAI · Open weight
44.8%
52
Nemotron 3 Super 120B A12BNVIDIA · Open weight
42.2%
53
Mercury 2Inception · Closed
39.6%
55
GPT-OSS 20BOpenAI · Open weight
36.3%
56
GPT-5.5 ProOpenAI · Closed
21.4%
58
Llama 4 ScoutMeta · Open weight
3.2%

The published PinchBench snapshot places anthropic/claude-opus-4.8-fast first at 93.5%. The third row is 3.0 points behind. The broader top-10 range is 6.5 points, so many of the published results sit in a relatively narrow band.

58 models have been evaluated on PinchBench. The benchmark falls in the Agentic category. This category carries a 22% weight in BenchLM.ai's overall scoring system. PinchBench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About PinchBench

Year

2026

Tasks

23 OpenClaw agent tasks

Format

Average success rate from official runs

Difficulty

Long-horizon agent workflows

PinchBench publishes official OpenClaw runs across 23 tasks and grades results with automated checks plus an LLM judge. BenchLM mirrors the public average-score view as a display-only benchmark.

BenchLM freshness & provenance

Version

PinchBench 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

FAQ

What does PinchBench measure?

An OpenClaw agent benchmark from Kilo that measures successful task completion across standardized real-world agent workflows.

Which model leads the published PinchBench snapshot?

anthropic/claude-opus-4.8-fast currently leads the published PinchBench snapshot with 93.5% average success rate. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on PinchBench?

58 AI models are included in BenchLM's mirrored PinchBench snapshot, based on the public leaderboard captured on 07/10/2026, 12:22 AM.

Last updated: 07/10/2026, 12:22 AM · mirrored from the public benchmark leaderboard

Choose a model with this week’s evidence

Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.

One email each week. Unsubscribe anytime.