Skip to main content

GBA-Eval

An agentic coding benchmark that asks models to build a Game Boy Advance emulator from scratch and grades emulator behavior against procedural, audio, and gameplay tests.

How BenchLM shows GBA-Eval

BenchLM mirrors the official GBA-Eval leaderboard snapshot graded on May 30, 2026. The benchmark asks coding agents to build a Game Boy Advance emulator and scores the result against 27 procedural, audio, and gameplay test cases.

GBA-Eval is display only on BenchLM. The source rows are agentic software-engineering runs with large token budgets and verifier-specific emulator tests, so BenchLM does not fold them into model-only weighted rankings.

14 agent rows27 emulator testsGBA emulator buildOfficial JSON feedDisplay only

Overall score on GBA-Eval — May 30, 2026

BenchLM mirrors the published overall score view for GBA-Eval. Claude Opus 4.8 leads the public snapshot at 70.9% , followed by GPT-5.5 (53.2%) and Claude Sonnet 4.6 (48.8%). BenchLM does not use these results to rank models overall.

14 modelsExternal benchmark mirrorsCurrentDisplay onlyUpdated May 30, 2026

The published GBA-Eval snapshot is tightly clustered at the top: Claude Opus 4.8 sits at 70.9%, while the third row is only 22.1 points behind. The broader top-10 spread is 70.0 points, so the benchmark still separates strong models even when the leaders cluster.

14 models have been evaluated on GBA-Eval. The benchmark falls in the External benchmark mirrors category. BenchLM tracks this category separately from its weighted global scoring system, so these results are best compared on the dedicated Korean benchmark views. GBA-Eval is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About GBA-Eval

Year

2026

Tasks

27 emulator test cases

Format

Overall emulator score

Difficulty

Long-horizon systems programming

GBA-Eval evaluates long-horizon coding agents by having them implement a working GBA emulator. The public leaderboard reports overall scores across 27 test cases with token usage and checkpoints preserved in the source JSON feed.

BenchLM freshness & provenance

Version

GBA-Eval 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.

Overall score table (14 models)

1
Claude Opus 4.8claude-opus-4-8
70.9%
2
GPT-5.5gpt-5.5
53.2%
3
Claude Sonnet 4.6claude-sonnet-4-6
48.8%
4
Claude Opus 4.6claude-opus-4-6
44.1%
5
Claude Opus 4.7claude-opus-4-7
43.8%
6
GPT-5.4gpt-5.4
31.6%
7
Gemini 3.5 Flashgoogle/gemini-3.5-flash
6.7%
8
Grok Build 0.1x-ai/grok-build-0.1
2.4%
9
MiniMax M3minimax/minimax-m3
0.9%
10
Kimi K2.6moonshotai/kimi-k2.6
0.9%
11
Gemini 3.1 Progemini-3.1-pro-preview
0.8%
12
Qwen 3.7 Maxqwen/qwen3.7-max
0.4%
13
GLM 5.1z-ai/glm-5.1
0.0%
14
MiniMax M2.7minimax/minimax-m2.7
0.0%

FAQ

What does GBA-Eval measure?

An agentic coding benchmark that asks models to build a Game Boy Advance emulator from scratch and grades emulator behavior against procedural, audio, and gameplay tests.

Which model leads the published GBA-Eval snapshot?

Claude Opus 4.8 currently leads the published GBA-Eval snapshot with 70.9% overall score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on GBA-Eval?

14 AI models are included in BenchLM's mirrored GBA-Eval snapshot, based on the public leaderboard captured on May 30, 2026.

Last updated: May 30, 2026 · mirrored from the public benchmark leaderboard

The AI models change fast. We track them for you.

For engineers, researchers, and the plain curious — a weekly brief on new models, ranking shifts, and pricing changes.

Free. No spam. Unsubscribe anytime.