agents benchmark
PinchBench
PinchBench evaluates coding agents on real-world agentic coding tasks, measuring both best-case and average performance across complex software engineering scenarios.
Updated Aug 7, 2026
Published models4
Registry coverage4
MetricScore
EvidenceB
PinchBench leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | NV | 90.0% | 100.0% | 4 | C | |
| 2 | XI | 81.2% | 66.7% | 4 | C | |
| 3 | XI | 81.0% | 33.3% | 4 | C | |
| 4 | ZA | 80.7% | 0.0% | 4 | C |
Score distribution
Top published rows on the benchmark's original scale.
PinchBench highlights
The top published results on this benchmark's own scale.
What is PinchBench?
Definition and scoring fields from the benchmark registry.
PinchBench evaluates coding agents on real-world agentic coding tasks, measuring both best-case and average performance across complex software engineering scenarios.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- PinchBench
- Modality
- text
- Primary category
- agents
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- pinchbench|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about PinchBench.
Which model scores highest on PinchBench?
Nemotron 3 Ultra (550B A55B) is currently ranked first with 90.0%.
What does PinchBench measure?
PinchBench evaluates coding agents on real-world agentic coding tasks, measuring both best-case and average performance across complex software engineering scenarios.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
4 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.