Get the Agent OS + join 4,000+ founders inside the AI Profit Boardroom → Join AIPB ($59/mo)

GoldieBench deep dive

Hy3's full benchmark breakdown.

Tencent's open-weights coder — Apache-2.0, cheap, beats GLM-5.1 on frontend in Tencent's blind eval. 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.

Data refreshed
2026-07-17
Scored tasks
7
Reading time
5 min
External sources
4

01 · The headline numbers

GoldieBench average
6.76/10
7 scored one-shot tasks
Board rank
#16
of 17 ranked frontier models
Task medals
0🥇 0🥈 0🥉
outright wins on shared briefs
Context
262,144-token context window. Open weights (Apache-2.0) on HuggingFace / ModelScope / GitHub; benched here via OpenRouter.
$0.14 / 1M input · $0.58 / 1M output

Provisional flag. Part of this model's run hit infrastructure failures that are being re-run; its average is still settling. Scores shown are real judged builds only.

02 · Every benchmark, bar by bar

All 7 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

“Clean multi-scene pipeline with animated count-up stats, gold/cyan cinematic palette, particle field and progress bar render correctly; but the stats appear left-clustered and off-center with only two of four visible mid-animation, feeling sparse rather than composed, keeping it solid-but-generic rather than a task winner.”

- the judge's verdict, unedited

“Clean render with a readable yellow hero car (wheels, cabin, taillight), colorful blocky city, working HUD/minimap and speed at 101km/h shows live play. Solid and functional but visually flat-lit and generic — buildings read as bare boxes and lighting is dim, keeping it below the top tier.”

- the judge's verdict, unedited

“Strong atmospheric snowy world with layered pines, soft shadows, snowfall, and clean HUD (health/stamina/compass/sword chip), but the hero reads as a stubby hooded blob with hidden face and no visible arms/legs, undercutting the flagship Skyrim-ranger fantasy.”

- the judge's verdict, unedited

04 · Where it struggles - quoted, not hidden

Boards that hide the weak rows are brochures. These are Hy3's lowest scored builds, verdicts unedited:

“HUD, minimap with dot-enemies, and a weapon read clearly, but the main view is a nearly featureless brown wall with no visible walls-vs-open geometry and no on-screen demon sprite, so the raycast world and monsters chasing you don't come across to the player. Solid UI polish can't rescue a flat, empty central frame.”

- the judge's verdict, unedited

“Renders cleanly with a detailed articulated skydiver (helmet, suit, arms, boots — not a bare capsule), clean HUD with altitude/distance/phase, and a lush jungle canopy of blobs with a visible river below; but the canopy overhead reads as a flat pink slab rather than a parachute, and the world feels more static/decorative than dynamic mid-drop, keeping it below the field's best.”

- the judge's verdict, unedited

“Clean HUD with working attitude indicator, heading tape, throttle/gear/score panels, and a decently modeled aircraft with wings, nav lights and prop; but the terrain reads as an empty green haze with no visible runway, trees or structures from this altitude, leaving the world flat and generic compared to the field's best.”

- the judge's verdict, unedited

“Strong dusk-city atmosphere with a readable blocky hero (cap, jacket trim, gun), streetlights, crosswalks, and a pedestrian, plus clean HUD (ammo, health bar, wanted stars, crosshair, controls). Weak points: the world is fairly barren mid-frame, no visible buildings-as-cover in this angle, and it reads more as a stylized diorama than the dense GTA street the brief implies, keeping it below the field's best.”

- the judge's verdict, unedited

05 · Category breakdown vs the whole field

Games (6 tasks)

Hy36.6
field avg7.1

Others (0 tasks)

field avg7.4

Pages (1 tasks)

Hy37.4
field avg7.5

Sims (0 tasks)

field avg6.7

Visuals (0 tasks)

field avg6.7

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

BenchmarkResultSource
SWE-bench Verified78.0 (GLM-5.2 leads at 84.2)hy.tencent.com
BrowseComp84.2 - leads the open fieldhy.tencent.com
MCP-Atlas79.1hy.tencent.com
LiveCodeBench-v634.86hy.tencent.com
Expert blind eval2.67/4 across 270 experts' real work tasks (GLM-5.1: 2.51/4)hy.tencent.com

From the source guides

MeasureResultSource guide
Expert blind eval2.67/4/hunyuan.tencent.com
Hallucination rate5.4%/hunyuan.tencent.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 Hy3 perform on GoldieBench?

Hy3 averages 6.76/10 across 7 scored one-shot build tasks, ranking #16 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 Hy3 best at?

Its strongest scored build is Aipbpromo at 7.4/10. By category it averages Game 6.6, Page 7.4 on the bench.

03How much does Hy3 cost?

Tencent Hunyuan 3 — open-weights under Apache-2.0, so free to self-host. On OpenRouter it is one of the cheapest capable coders: ~$0.14/M in, $0.58/M out (1 RMB / 4 RMB). Upstream can be slow (30-90s to first token), but per-token cost is negligible.

Source ledger

  1. 01Official vendor site: openrouter.ai/tencent/hy3openrouter.ai
  2. 02SWE-bench Verifiedhy.tencent.com
  3. 03Expert blind evalagentos.guide
  4. 04Julian's guide: /hy3-agent-osagentos.guide

Every model's breakdown

The same stack Julian uses

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.

4,000+founders
258documented wins
38countries
$59/momonthly