Living benchmarkVersion 0.1
AI Go Problem Creation Benchmark
A reproducible evaluation of whether AI models can author original, legal, and teachable life-and-death Go problems on 19×19 boards in SGF. Every result links back to its generated position, solution tree, automated checks, and human review.
- Models evaluated
- 11
- Runs indexed
- 15
- Evaluation slots
- 150
- Human reviews
- 111
- Latest update
- Aug 08, 2026 UTC · indexed artifacts
BENCHMARK TRAJECTORY
All LLMs By Release Date
Each model is plotted once by public release date, using its best human-reviewed run and a difficulty-capped score out of 10.
Score out of 10
0246810Apr 10, 2026May 12Jun 14Jul 16Aug 17, 2026
Model release date
Current results
Model summary
Each row uses that model’s best human-reviewed run. Select a model to inspect its generated files and gate-by-gate evidence.
| Model | Human score | Automated gate | Structural | Runs | Latest best |
|---|---|---|---|---|---|
| Claude Fable 5Anthropic | 2/102.0 / 10 | 10 / 10 | 10 / 10 | 1 | Aug 08, 2026 |
| Claude Opus 5Anthropic | 2/102.0 / 10 | 10 / 10 | 10 / 10 | 2 | Aug 07, 2026 |
| GPT-5.6 SolOpenAI | 1/101.0 / 10 | 10 / 10 | 10 / 10 | 4 | Aug 07, 2026 |
| Claude Opus 4.8Anthropic | 0/100.0 / 10 | 10 / 10 | 10 / 10 | 1 | Aug 08, 2026 |
| DeepSeek V4 Flash 0731DeepSeek | 0/100.0 / 10 | 10 / 10 | 10 / 10 | 1 | Aug 07, 2026 |
| GPT-5.5 (Codex)OpenAI | 0/100.0 / 10 | 10 / 10 | 10 / 10 | 1 | Aug 08, 2026 |
| GPT-5.6 LunaOpenAI | 0/100.0 / 10 | 10 / 10 | 10 / 10 | 1 | Aug 07, 2026 |
| Grok 4.5xAI | 0/100.0 / 10 | 10 / 10 | 10 / 10 | 1 | Aug 07, 2026 |
| Muse Spark 1.2Meta | 0/100.0 / 10 | 10 / 10 | 10 / 10 | 1 | Aug 08, 2026 |
| Kimi K3Kimi | 0/100.0 / 10 | 0 / 10 | 0 / 10 | 1 | Aug 08, 2026 |
| Qwen3.8-MaxQwen | 0/100.0 / 10 | 0 / 10 | 1 / 10 | 1 | Aug 08, 2026 |
Run log
Latest benchmark activity
| Started | Model | Effort | Structural | Automated gate | Reviewed | Status |
|---|---|---|---|---|---|---|
| Aug 08, 2026, 15:25 | Claude Fable 5Anthropic | low | 10/10 | 10/10 | 10/10 | reviewed |
| Aug 08, 2026, 04:14 | GPT-5.6 SolOpenAI | max | 10/10 | 10/10 | 10/10 | reviewed |
| Aug 08, 2026, 02:09 | Qwen3.8-MaxQwen | default | 1/10 | 0/10 | 1/10 | failed |
| Aug 08, 2026, 01:47 | Muse Spark 1.2Meta | default | 10/10 | 10/10 | 10/10 | reviewed |
| Aug 08, 2026, 01:44 | Kimi K3Kimi | default | 0/10 | 0/10 | 0/10 | failed |
| Aug 08, 2026, 01:40 | GPT-5.5 (Codex)OpenAI | xhigh | 10/10 | 10/10 | 10/10 | reviewed |
Audit the benchmark
Methods and source material
Results are useful only when the task, artifacts, and review criteria remain inspectable.