GoldieBench deep dive
Qwen 3.7's full benchmark breakdown.
Multilingual open-weights — strong on Chinese reasoning. 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 47 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
“The closest test. All three shipped a real, juicy game. Opus's breakout had the most game-feel (particle bursts + live combo). Qwen's neon breakout is clean and vibrant. GLM went its own way with fullscreen asteroids. Genuinely hard to separate.”
“GLM and Opus both produced premium gradient 'Intelligence, reimagined / distilled' keynote heroes — basically a tie. Qwen's is clean and well-built (proper nav + three feature cards) but the headline ('Built for the next generation of builders') lands flatter than the gradient heroes.”
04 · Where it struggles - quoted, not hidden
Boards that hide the weak rows are brochures. These are Qwen 3.7's lowest scored builds, verdicts unedited:
“0KB · no animation detected on load · TIMEOUT”
“14KB · animation runs but no input response · webgl”
“6KB · animation runs but no input response · webgl, rAF”
“11KB · animation runs but no input response · plain”
05 · Category breakdown vs the whole field
Games (23 tasks)
Others (0 tasks)
Pages (3 tasks)
Sims (12 tasks)
Visuals (9 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.
Vendor-published numbers
| Benchmark | Result | Source |
|---|---|---|
| SWE-bench Pro | 60.6% | amitray.com |
| SWE-bench Verified | 80.4% | amitray.com |
| Terminal-Bench 2.0 Terminus | 69.7 | amitray.com |
| MCP-Atlas | 76.4 | amitray.com |
| Autonomy claim | runs up to 35 hours autonomously; supports external harnesses incl. Claude Code | venturebeat.com |
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 Qwen 3.7 perform on GoldieBench?
Qwen 3.7 averages 7.00/10 across 47 scored one-shot build tasks, ranking #15 of 17 ranked frontier models, with 0 task golds, 0 silvers and 0 bronzes. Every score is a real judged build you can open and play on this site.
02What is Qwen 3.7 best at?
Its strongest scored build is Arcade at 8.0/10. By category it averages Game 7.1, Page 7.0, Sim 6.9, Visual 6.9 on the bench.
03How much does Qwen 3.7 cost?
Alibaba's open-weights release — downloadable from Hugging Face, runnable locally or via Alibaba Cloud's free tier for individuals.
Source ledger
- 01Official vendor site: chat.qwen.aichat.qwen.ai
- 02SWE-bench Proamitray.com
- 03Autonomy claimventurebeat.com
- 04Julian's guide: /glm-vs-qwen-vs-opusagentos.guide
- 05Julian's guide: /qwen-hermesagentos.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.