GoldieBench deep dive
MiniMax M3's full benchmark breakdown.
1M-context frontier model at $0.30/M tokens — cheapest big-context model on the bench. 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
“34KB frozen open world — snowy mountains, pines, flying dragon, full HUD.”
“41KB Nordic crypt with torch-lit corridors, chasing skeletons, boss room.”
“59KB third-person arcade racer. Banking turns, speed boost, drift, lap timer.”
“31KB Gray-Scott shader with click-to-seed.”
04 · Where it struggles - quoted, not hidden
Boards that hide the weak rows are brochures. These are MiniMax M3's lowest scored builds, verdicts unedited:
“Minimal 1KB plasma. Brief is barely met.”
“Canvas-2D raycaster — WASD walking, textured walls, distance fog.”
“Minimal gravitational-lens shader.”
“3D tunnel flythrough with distorted starfield.”
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 | 59.0% - vendor-run with agent scaffolding | www.marktechpost.com |
| Terminal-Bench 2.1 | 66.0% | www.marktechpost.com |
| BrowseComp | 83.5 - above Opus 4.7's 79.3 | felloai.com |
| MMMU-Pro (multimodal) | 78.1% | felloai.com |
Independent evaluations
| Benchmark | Result | Source |
|---|---|---|
| Verification status | several results ran on MiniMax's own infrastructure with scaffolding; independent verification pending | www.techtimes.com |
From the source guides
| Measure | Result | Source guide |
|---|---|---|
| Context window | 1,048,576 tokens | /openrouter |
| Per-token cost | $0.30 / M input · $1.50 / M output | /openrouter |
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 MiniMax M3 perform on GoldieBench?
MiniMax M3 averages 7.97/10 across 47 scored one-shot build tasks, ranking #7 of 17 ranked frontier models, with 2 task golds, 3 silvers and 7 bronzes. Every score is a real judged build you can open and play on this site.
02What is MiniMax M3 best at?
Its strongest scored build is Dragonrealm at 9.0/10. By category it averages Game 8.2, Page 8.2, Sim 7.9, Visual 7.4 on the bench.
03How much does MiniMax M3 cost?
MiniMax M3 is the cheapest 1M-context frontier model on the bench — roughly 1/200th the per-call cost of OpenRouter Fusion and 1/30th of Claude Opus 4.8. Designed for high-volume agent workloads where context length matters but per-call budget is tight.
Source ledger
- 01Official vendor site: minimax.iowww.minimax.io
- 02SWE-bench Prowww.marktechpost.com
- 03BrowseCompfelloai.com
- 04Verification statuswww.techtimes.com
- 05Context windowagentos.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.