GoldieBench deep dive
Fugu Mini's full benchmark breakdown.
Fugu's fast mini variant — single model, no panel, ~3 min per build. Every number below is either one of my own judged one-shot builds - playable on this site - or an outside result with its source linked. Nothing display-only, nothing vibes.
01 · The headline numbers
02 · Every benchmark, bar by bar
All 36 scored tasks, best first. Hover for the judge's comment; click through for every model's build on that task. Or overlay other models in the interactive graphs.
03 · Where it wins - with the judge's own words
“Juicy browser game. Smoke-test PASS with 55% pixel diff — most reactive build in the sweep.”
“Mini gap-fill (round 2) — Tron procedural terrain. Smoke-test PASS (8.6% diff — strong motion).”
04 · Where it struggles - quoted, not hidden
Boards that hide the weak rows are brochures. These are Fugu Mini's lowest scored builds, verdicts unedited:
“Dogfight scene structurally complete but smoke-test shows MAYBE-STATIC (<0.5% pixel diff after input) — possible the plane isn't taking keys reliably.”
“Mini gap-fill (round 2) — desktop OS shell. Smoke-test MAYBE (0.0% diff) — static until an app is opened; flagged for manual verification.”
“Top-down RPG. Smoke-test MAYBE-STATIC — character may be present but minimal response to keys.”
“Spiral galaxy. Smoke-test MAYBE-STATIC (<0.5% pixel diff) — slow rotation may be present without strong input response.”
05 · Category breakdown vs the whole field
Games (14 tasks)
Others (0 tasks)
Pages (2 tasks)
Sims (12 tasks)
Visuals (8 tasks)
06 · Outside signals - every row sourced
My bench measures one thing: judged one-shot builds. These outside rows measure other things - kept separate, never blended into the GoldieBench average, every value linked to where it comes from.
No externally sourced rows tracked for this model yet - the GoldieBench scores above are the record.
07 · Head-to-head records
08 · Methodology + honest limits
Every GoldieBench score is a one-shot, single-file build from an identical prompt - no retries, no hand-fixing - rendered for real, screenshotted, and scored 0-10 by one judge model on one rubric across the whole field. Failures score as failures. What this bench does NOT measure: multi-turn agent work, long-context recall, or API latency - that is what the sourced outside rows are for. Full method: /methodology.
Frequently asked questions
01How does Fugu Mini perform on GoldieBench?
Fugu Mini averages 7.75/10 across 36 scored one-shot build tasks, ranking #11 of 17 ranked frontier models, with 2 task golds, 0 silvers and 3 bronzes. Every score is a real judged build you can open and play on this site.
02What is Fugu Mini best at?
Its strongest scored build is Game at 9.0/10. By category it averages Game 7.8, Page 7.0, Sim 7.7, Visual 8.0 on the bench.
03How much does Fugu Mini cost?
The non-Ultra `fugu` model on Sakana's API. Sakana describes it as 'Fast mini model optimized for low latency yet high quality responses.' Crucially: zero orchestration tokens per call (vs Ultra's panel of thousands). Returns in ~3 min instead of 6-15 min and doesn't time out on heavy prompts.
Source ledger
- 01Official vendor site: sakana.aisakana.ai
- 02Julian's guide: /sakana-fugu-vs-fusionagentos.guide
Every model's breakdown
Run this stack yourself.
Every demo on this bench was built inside the Agent Operating System — one prompt, one shot, single HTML file out. The Agent OS, the prompts, the templates, the weekly walkthroughs and 4,000+ founders shipping with it every day all live inside the AI Profit Boardroom.