BenchLeader
Menu
Best LLMsChoose a modelCoding plansLeaderboardsModelsBenchmarksProvidersReleases7

ProgramBench leaderboard

As of 6 Nov 2026, the top score on ProgramBench is 18.5% by Claude Opus 5.5 (Anthropic), ahead of Claude Fable 5.1 at 7.0%, across 64 model configurations with a published result (Vals AI, result dated 5 Nov 2026). Given a binary, the model must recover working source that reproduces its behaviour, checked by tests. ProgramBench is shown for reference only and does not count toward the BenchLeader Index.

Key facts (as of 6 Nov 2026)

  • Claude Opus 5.5 holds the top ProgramBench score at 18.5% (Vals AI, 5 Nov 2026).
  • Claude Fable 5.1 is second on ProgramBench at 7.0% (Vals AI, 5 Nov 2026).
  • ProgramBench: Rebuilding programs from compiled binaries. Run by Vals AI (Vals AI).
  • 64 model configurations have a published ProgramBench result on BenchLeader (BenchLeader, 6 Nov 2026).
  • ProgramBench does not feed the BenchLeader Index; it is shown for reference (BenchLeader methodology).

Rebuilding programs from compiled binaries. Run by Vals AI.

Published by
Vals AI
Category
Coding
Index weight
Reference only
Models
64
Data as of
6 Nov 2026

Vals AI (vals.ai).

What the test looks like

Given a binary, the model must recover working source that reproduces its behaviour, checked by tests.

How it is scored

Percent of programs rebuilt, run by Vals AI.

What to keep in mind

Reverse engineering is niche; scores are low for most models.

Top 15
  1. 1Claude Opus 5.518.5%
  2. 2Claude Fable 5.17.0%
  3. 3Claude Sonnet 5.56.5%
  4. 4GPT-6 Astra (max)5.5%
  5. 5GPT-6.1 Sol (max)3.0%
  6. 6Claude Opus 53.0%
  7. 7Gemini 4 Argon (high)2.5%
  8. 8Muse Spark 1.3 (max)2.5%
  9. 9Claude Haiku 5.52.5%
  10. 10Claude Fable 52.0%
  11. 11GPT-6 Sol (max)2.0%
  12. 12Kimi K32.0%
  13. 13GPT-5.6 Sol (max)1.5%
  14. 14GLM 5.3 (max)1.5%
  15. 15Gemini 3.8 Flash (high)1.0%
64 of 64
#
1Claude Opus 5.518.5%63.4$68.052026-11-05
2Claude Fable 5.17.0%66.4$58.522026-11-05
3Claude Sonnet 5.56.5%–$35.862026-11-05
4GPT-6 Astramax5.5%67.8$11.572026-11-05
5GPT-6.1 Solmax3.0%67.3$2.872026-11-05
6Claude Opus 53.0%65.2$60.292026-11-05
7Gemini 4 Argonhigh2.5%66.0$17.662026-11-05
8Muse Spark 1.3max2.5%62.0$54.292026-11-05
9Claude Haiku 5.52.5%–$6.832026-11-05
10Claude Fable 52.0%66.5$75.682026-11-05
11GPT-6 Solmax2.0%63.3$11.672026-11-05
12Kimi K32.0%59.1$70.482026-11-05
13GPT-5.6 Solmax1.5%65.4$15.292026-11-05
14GLM 5.3max1.5%59.0$21.962026-11-05
15Gemini 3.8 Flashhigh1.0%62.0$10.932026-11-05
16Grok 4.6high1.0%60.5$16.762026-11-05
17Claude Opus 4.81.0%60.0$31.272026-11-05
18GPT-5.5xhigh0.5%64.3$6.952026-11-05
19Muse Wonder Fullmax0.5%62.9$28.652026-11-05
20Muse Spark 1.2xhigh0.5%61.7$2.472026-11-05
21GPT-5.6 Terramax0.5%60.7$5.382026-11-05
22GLM 5.2max0.5%59.5$12.582026-11-05
23DeepSeek V4.1 Flashmax0.5%59.0$0.8972026-11-05
24GPT-5.4high0.5%58.8$2.272026-11-05
25Muse Spark 1.3xhigh0.5%58.5$11.562026-11-05
26Grok 4.7xhigh0.5%58.3$46.492026-11-05
27Claude Sonnet 4.60.5%58.2–2026-11-05
28NewInkling 2 Small0.5%57.4$8.462026-11-05
29MiMo-V2.6-Pro0.5%57.3$0.7272026-11-05
30GPT-6 Lunamax0.5%56.7$0.1812026-11-05
31MiMo-V2.6-Flash0.5%55.0$0.4712026-11-05
32Inkling Small0.5%52.1$0.4792026-11-05
33GPT-5.4xhigh0.0%63.2–2026-11-05
34Claude Opus 4.70.0%63.1–2026-11-05
35Gemini 3.5 Flashhigh0.0%62.5–2026-11-05
36Gemini 3.7 Flashhigh0.0%62.2$7.132026-11-05
37Gemini 3.6 Flashhigh0.0%61.2$5.922026-11-05
38Gemini 3.1 Prohigh0.0%60.1$2.632026-11-05
39Muse Spark 1.1xhigh0.0%59.5$13.062026-11-05
40Qwen3.6 Plus0.0%59.1–2026-11-05
41Qwen3.8 Maxmax0.0%58.8$12.362026-11-05
42Gemini 3 Flashhigh0.0%58.3$0.4402026-11-05
43DeepSeek V4 Flashhigh0.0%57.3$1.292026-11-05
44DeepSeek V4 Promax0.0%57.0$3.412026-11-05
45Grok 4.5high0.0%56.9$4.122026-11-05
46Grok 4.3high0.0%56.6$0.3562026-11-05
47MiniMax M2.70.0%56.5–2026-11-05
48GPT-5.6 Lunamax0.0%56.3$0.9432026-11-05
49GPT-5.4 Minixhigh0.0%56.0–2026-11-05
50Claude Sonnet 50.0%55.9$24.362026-11-05
51GLM 5.10.0%55.7–2026-11-05
52Hy40.0%55.7$12.812026-11-05
53Nemotron 3 Ultra 550B A55B0.0%55.5–2026-11-05
54Kimi K2.60.0%55.2–2026-11-05
55Kimi K2.7 Code0.0%54.2$3.192026-11-05
56GLM 5.3 Flashmax0.0%54.2$1.602026-11-05
57Qwen3.8 27Bxhigh0.0%52.5$17.992026-11-05
58Gemini 3.1 Flash Litehigh0.0%48.5$0.0642026-11-05
59Gemini 3.5 Flash Litehigh0.0%47.9$0.3372026-11-05
60Inkling0.0%47.5$4.142026-11-05

More coding benchmarks · All LLM benchmarks

Related benchmarks

Top head-to-heads on ProgramBench

Cite as: BenchLeader, “ProgramBench leaderboard”, data as of 6 Nov 2026, https://www.benchleader.com/benchmarks/vals_programbench.

ProgramBench: questions

What does ProgramBench measure?
Given a binary, the model must recover working source that reproduces its behaviour, checked by tests. Scores are reported in percent of tasks solved; higher is better.
What is the top score on ProgramBench?
As of 6 Nov 2026, the highest score on ProgramBench is 18.5% by Claude Opus 5.5 (result dated 5 Nov 2026), ahead of Claude Fable 5.1 at 7.0%. Scores come from Vals AI and are re-read daily.
Which AI model leads ProgramBench?
Claude Opus 5.5 leads ProgramBench with 18.5% as of 6 Nov 2026 (result dated 5 Nov 2026), ahead of Claude Fable 5.1 at 7.0%.
How many models have ProgramBench results?
64 model configurations have a ProgramBench result on BenchLeader, all taken from Vals AI.
Who runs ProgramBench and how often is it updated?
ProgramBench is published by Vals AI. BenchLeader re-reads the published results once a day and records the date each result was published.
Does ProgramBench count toward the BenchLeader Index?
No. ProgramBench is shown for reference but left out of the composite index.