Benchmark Almanac

GBAEval

Claude Opus 5 79.6% leads since July 2026
Not in model cards

Agents get 24 hours to write a Game Boy Advance emulator in Rust for WebAssembly. Replay, hardware, and audio tests compare it with Mesen2.

Mechanize describes it as one test case rather than a full benchmark, so each score rests on a single task.

Made by
Mechanize
Released
May 2026
Kind
Building whole projects
Level
Research frontier
Scored
Checked automatically
Answers
Code
Tasks
All public
Strongest model tested
Claude Opus 5 #10 on the capability index
Last model added
July 2026
Scores refreshed
November 12th, 2026
21
Models scored on its own leaderboard

Claude Opus 5 leads with 79.6% among models released by July 2026

Best score per model from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best.

The newest model with a score here came out in July 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

21 models. Scroll sideways for the rest.

  1. 79.6 Claude Opus 5
  2. 74.5 Claude Fable 5
  3. 70.9 Claude Opus 4.8
  4. 65.4 Grok 4.5
  5. 65.3 Claude Sonnet 5
  6. 53.2 GPT-5.5
  7. 52.6 GPT-5.6 Sol
  8. 48.8 Claude Sonnet 4.6
  9. 48.4 Kimi K3
  10. 45.1 GPT-5.4
  11. 44.1 Claude Opus 4.6
  12. 43.8 Claude Opus 4.7
  13. 7.92 Muse Spark 1.1
  14. 6.70 Gemini 3.5 Flash
  15. 2.38 Grok Build 0.1
  16. 0.91 MiniMax-M3
  17. 0.86 Kimi K2.6
  18. 0.85 Gemini 3.1 Pro
  19. 0.85 Kimi K2.7 Code
  20. 0.41 Qwen3.7-Max
  21. 0.02 GLM-5.2
  1. 1 Claude Opus 5 79.6%
  2. 2 Claude Fable 5 74.5%
  3. 3 Claude Opus 4.8 70.9%
  4. 4 Grok 4.5 65.4%
  5. 5 Claude Sonnet 5 65.3%
  6. 6 GPT-5.5 53.2%
  7. 7 GPT-5.6 Sol 52.6%
  8. 8 Claude Sonnet 4.6 48.8%
  9. 9 Kimi K3 48.4%
  10. 10 GPT-5.4 45.1%
  11. 11 Claude Opus 4.6 44.1%
  12. 12 Claude Opus 4.7 43.8%
  13. 13 Muse Spark 1.1 7.92%
  14. 14 Gemini 3.5 Flash 6.70%
  15. 15 Grok Build 0.1 2.38%
  16. 16 MiniMax-M3 0.91%
  17. 17 Kimi K2.6 0.86%
  18. 18 Gemini 3.1 Pro 0.85%
  19. 19 Kimi K2.7 Code 0.85%
  20. 20 Qwen3.7-Max 0.41%
  21. 21 GLM-5.2 0.02%

Source: Official GBAEval leaderboard

#ModelScoreSettingDateSource
1 Claude Opus 5 79.6% Jul 2026 Official leaderboard
2 Claude Fable 5 74.5% Jun 2026 Official leaderboard
3 Claude Opus 4.8 70.9% May 2026 Official leaderboard
4 Grok 4.5 65.4% Jul 2026 Official leaderboard
5 Claude Sonnet 5 65.3% Jun 2026 Official leaderboard
6 GPT-5.5 53.2% Apr 2026 Official leaderboard
7 GPT-5.6 Sol 52.6% Codex Jul 2026 Official leaderboard
8 Claude Sonnet 4.6 48.8% Feb 2026 Official leaderboard
9 Kimi K3 48.4% Jul 2026 Official leaderboard
10 GPT-5.4 45.1% Mar 2026 Official leaderboard
11 Claude Opus 4.6 44.1% Feb 2026 Official leaderboard
12 Claude Opus 4.7 43.8% Apr 2026 Official leaderboard
13 Muse Spark 1.1 7.92% goose Jul 2026 Official leaderboard
14 Gemini 3.5 Flash 6.70% goose May 2026 Official leaderboard
15 Grok Build 0.1 2.38% May 2026 Official leaderboard
16 MiniMax-M3 0.91% goose Jun 2026 Official leaderboard
17 Kimi K2.6 0.86% goose Apr 2026 Official leaderboard
18 Gemini 3.1 Pro 0.85% Feb 2026 Official leaderboard
19 Kimi K2.7 Code 0.85% goose Jun 2026 Official leaderboard
20 Qwen3.7-Max 0.41% goose May 2026 Official leaderboard
21 GLM-5.2 0.02% goose Jun 2026 Official leaderboard

4 models have held number one on the official leaderboard since its release in May 2026

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI Anthropic
Own leaderboard Not tested since Jul 2026
Claude Opus 5 79.6%

Source: Official GBAEval leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5 Claude Fable 5 GPT-5.5 Claude Sonnet 4.6 Claude Opus 4.6 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Claude Opus 5 Claude Opus 4.8

Source: Official GBAEval leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Coding

All 282 in Coding
  • SWE-bench Verified In use 500 human-validated GitHub issues from 12 Python repos. Agent writes a patch that must pass hidden tests. Claude Opus 4.7 83.5%
  • LiveCodeBench v6 In use Competitive programming problems collected after model cutoffs from LeetCode, AtCoder, Codeforces. Scored pass@1 on hidden tests. DeepSeek-V4-Pro 93.5%
  • SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Fable 5.2 96.6%
  • DeepSWE In use 113 original long-horizon engineering tasks across 91 repos and five languages, written from scratch. Hand-written verifiers accept alternative solutions. GPT-6 Astra 74.1%
  • SciCode In use Scientist-curated research coding problems split into subproblems across natural sciences. Code checked against test cases. Seed2.1 Pro 59.8%
  • Codeforces rating In use Model solves recent Codeforces contest problems. An Elo-style rating is computed from simulated contest performance. DeepSeek V4.1 Flash 3,471