agents benchmark
Program Bench Leaderboard
Program Bench evaluates code-generation agents by asking them to recreate a program's behavior from only a compiled binary and documentation. It spans 200 tasks from small CLI tools to large systems such as FFmpeg and SQLite, with submissions judged against more than 248,000 fuzz-generated behavioral tests.
Updated Aug 17, 2026
Models6
Model coverage6
MetricScore
EvidenceB
Program Bench Ranking
Higher score ranks better on this benchmark.
Sort by
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelMA | Score77.8% | Percentile100.0% | Participants6 | EvidenceC | Evaluated |
| Rank02 | ModelZA | Score63.7% | Percentile80.0% | Participants6 | EvidenceC | Evaluated |
| Rank03 | ModelMA | Score53.6% | Percentile60.0% | Participants6 | EvidenceC | Evaluated |
| Rank04 | ModelBY | Score50.3% | Percentile40.0% | Participants6 | EvidenceC | Evaluated |
| Rank05 | ModelBY | Score49.4% | Percentile20.0% | Participants6 | EvidenceC | Evaluated |
| Rank06 | ModelZA | Score19.0% | Percentile0.0% | Participants6 | EvidenceC | Evaluated |
Program Bench Highlights
The leading models and scores on this benchmark.
Program Bench Score Distribution
A closer view of the leading scores on this benchmark.
The Top AI Models for Program Bench
The first five results on this benchmark, with official price and output speed added where the model identity can be matched.
Ranking basisThis program bench AI model leaderboard uses descending score in the benchmark's original unit. The leaderboard ranking keeps matched price and speed data separate from benchmark evidence.
- 01MAKimi K3Moonshot AI
- Score
- 77.8%
- Price
- $3.0 input / $15 output per 1M tokens
- Speed
- Up to 26 tok/s via Fireworks
Strengths
- Ranks #1 of 6 compared models
- 100th percentile on this benchmark
- C evidence result
Considerations
- This result measures Program Bench, not total model capability
- 02ZAGLM-5.2Zhipu AI
- Score
- 63.7%
- Price
- $1.4 input / $4.4 output per 1M tokens
Strengths
- Ranks #2 of 6 compared models
- 80th percentile on this benchmark
- C evidence result
Considerations
- This result measures Program Bench, not total model capability
- 03MAKimi K2.7 CodeMoonshot AI
- Score
- 53.6%
- Price
- $0.95 input / $4.0 output per 1M tokens
- Speed
- Up to 36 tok/s via Novita
Strengths
- Ranks #3 of 6 compared models
- 60th percentile on this benchmark
- C evidence result
Considerations
- This result measures Program Bench, not total model capability
- 04BYSeed 2.1 ProByteDance
- Score
- 50.3%
Strengths
- Ranks #4 of 6 compared models
- 40th percentile on this benchmark
- C evidence result
Considerations
- This result measures Program Bench, not total model capability
- 05BYSeed 2.1 TurboByteDance
- Score
- 49.4%
Strengths
- Ranks #5 of 6 compared models
- 20th percentile on this benchmark
- C evidence result
Considerations
- This result measures Program Bench, not total model capability
Selection summary
Best AI Models for Program Bench
Kimi K3 currently leads Program Bench with 77.8%. It is the top model on this specific benchmark, while the best LLM for the broader task should also be checked against other benchmarks, price and runtime.
Use this leaderboard with the supporting benchmark results and coverage details above. A leaderboard position summarizes the selected ranking signal; it does not replace workload-specific testing.
What is Program Bench?
What Program Bench measures and how its scores work.
Program Bench evaluates code-generation agents by asking them to recreate a program's behavior from only a compiled binary and documentation. It spans 200 tasks from small CLI tools to large systems such as FFmpeg and SQLite, with submissions judged against more than 248,000 fuzz-generated behavioral tests.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- Program Bench
- Modality
- text
- Primary category
- agents
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- program-bench|llm-stats-current
Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about Program Bench.
Which model scores highest on Program Bench?
Kimi K3 is currently ranked first with 77.8%.
What does Program Bench measure?
Program Bench evaluates code-generation agents by asking them to recreate a program's behavior from only a compiled binary and documentation. It spans 200 tasks from small CLI tools to large systems such as FFmpeg and SQLite, with submissions judged against more than 248,000 fuzz-generated behavioral tests.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
6 model results are currently shown.
Does this benchmark affect the overall score?
Yes. This benchmark can contribute to the current LLMBoard capability score.