Tencent Hunyuan

Hy3

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

Context262,144-token context window. Open weights (Apache-2.0) on HuggingFace / ModelScope / GitHub; benched here via OpenRouter.
Pricing$0.14 / 1M input · $0.58 / 1M output
Tasks tested7
Avg score7.13/10 average
Medals🥇0 🥈1 🥉0
Release2026-07-06
Official vendor source
Hy3 is built by Tencent Hunyuan — see the vendor's own product page, pricing, and docs at openrouter.ai/tencent/hy3.
Visit openrouter.ai/tencent/hy3 →

Reference benchmarks for Hy3

These are external benchmarks I pulled from the source comparison guides on agentos.guide — SWE-bench Verified, DRACO, Kilo plan rubric, build-time measurements, vendor-reported coding scores. They are not goldiebench medal scores (those come only from same-prompt one-shot creative coding tasks in the matrix). I surface them here so the spec sheet for Hy3 is honest about what's measured.

Expert blind eval
2.67/4
Hallucination rate
5.4%

What is Hy3?

Hy3 is the Tencent Hunyuan frontier model with a 262,144-token context window. Open weights (Apache-2.0) on HuggingFace / ModelScope / GitHub; benched here via OpenRouter. context window, released 2026-07-06. Tagline: Tencent's open-weights coder — Apache-2.0, cheap, beats GLM-5.1 on frontend in Tencent's blind eval.. Official source: openrouter.ai/tencent/hy3.

Pricing detail. 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.

How I use it inside the Agent OS. Wired into the Agent OS as the 'Hy3 Coder' tab (chat + live preview + workspace) via OpenRouter. Bench built one-shot on the same prompts as the field; weak builds iterated by Hy3 itself (the model fixes its own builds).

What I built with Hy3

Every model on Goldie Bench gets the same fixed prompt set — one shot, single HTML file out — and I score the result 0–10 inside the Agent Operating System. Here's what Hy3 shipped on the bench: 7 one-shot demos across 262,144-token context window. Open weights (Apache-2.0) on HuggingFace / ModelScope / GitHub; benched here via OpenRouter. of context. Of those, 7 are scored against the field with my honest 0–10 from the source guides at agentos.guide.

Strengths

  • Apache-2.0 open weights — self-host free, no lock-in
  • Tencent's 270-expert blind eval: 2.67/4 vs GLM-5.1's 2.51, strongest on frontend / data / CI-CD
  • Hallucination rate cut 12.5% → 5.4%; stable tool-calls across scaffoldings (<4% SWE-Bench variance)

Trade-offs

  • Slow upstream on OpenRouter (30-90s per build) — fine for one-shots, sluggish for tight loops
  • One-shot game builds can under-render (flat raycaster walls, unlit 3D) without an iterate pass

Best for

  • Cost-sensitive coding + frontend design where open weights matter
  • Self-hosters who want an Apache-2.0 model they fully own
  • Anyone wiring a cheap capable coder into a live build panel (Agent OS Hy3 Coder tab)

Every demo by Hy3

7 live demos, sorted by category. Click any tile to play the actual one-shot result. Verdicts and 0–10 scores are pulled from the source guides where I posted them publicly.

Dragonrealm▶ LIVE
Dragonrealm
Game
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.
Doom▶ LIVE
Doom
Game
Demon sprites with horns/eyes/teeth read clearly and the HUD, minimap and vignette are polished, but there's no visible wall structure — the whole frame is a flat brown void with sprites floating on it, so the raycaster/maze isn't reading as a 3D world to the player.
Gtafoot▶ LIVE
Gtafoot
Game
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.
Gtadrive▶ LIVE
Gtadrive
Game
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.
Aipbpromo▶ LIVE
Aipbpromo
Page
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.
Parachute▶ LIVE
Parachute
Game
Clean HUD (altitude, dist-to-target arrow, phase) and a nicely modeled skydiver with helmet visor, orange suit and boots reads well against drifting clouds. But mid-freefall the frame is mostly empty sky with no jungle/ground/clearing in view, so it feels sparse and generic rather than dramatic; solid but not a task-topper.
Flightsim▶ LIVE
Flightsim 🥈
Game
Clean readable HUD (SPD/ALT/VS, throttle bar, attitude ball, heading tape) plus a well-modeled shaded aircraft with red/blue wingtip lights, spinning prop, runway, tower, trees and lake — polished and clearly on-brief. Falls just short of the field's best: the attitude indicator is a flat ellipse and the runway/terrain read a bit sparse, keeping it strong-shippable rather than a task winner.
every demo, in a grid · click any one to play

Compare Hy3 against every other model

Every head-to-head featuring Hy3. Verdicts shown for scored pairs.

Hy3 vs Fusion
Fusion leads 7–0
Hy3 vs Hermes MoA
Hy3 leads 4–3
Hy3 vs Claude Fable 5
Claude Fable 5 leads 6–0
Hy3 vs Grok
Grok leads 4–1
Hy3 vs MiniMax M3
MiniMax M3 leads 6–0
Hy3 vs Fugu Ultra
Tied 1–1
Hy3 vs GLM-5.2
GLM-5.2 leads 6–0
Hy3 vs Fugu Mini
Fugu Mini leads 2–0
Hy3 vs Opus 4.8
Opus 4.8 leads 4–2
Hy3 vs Kimi K2.7
Kimi K2.7 leads 4–1
Hy3 vs Qwable 5 27B Coder
Qwable 5 27B Coder leads 2–0
Hy3 vs Claude Sonnet 5
Hy3 leads 4–3
Hy3 vs Qwen 3.7
Qwen 3.7 leads 4–3
Hy3 vs Agents-A1
Hy3 leads 2–0
Hy3 vs Gemma 4 12B · MLX
Hy3 leads 2–0
Hy3 vs Laguna XS 2.1
Hy3 leads 2–0
Hy3 vs Qwythos 9B
Hy3 leads 2–0
Hy3 vs LongCat-2.0
LongCat-2.0 leads 1–0
Hy3 vs Gemma-4 12B Coder
2 shared tasks · unscored
Hy3 vs Kimi K2.7 · Fast
7 shared tasks · unscored
Hy3 vs Kimi K2.7 · No-Think
7 shared tasks · unscored
Hy3 vs Kimi K2.7 · Quality
7 shared tasks · unscored
Hy3 vs Ornith 1.0
2 shared tasks · unscored
Hy3 vs Claude Mythos 5
Reference-only
Hy3 vs Kilo Code
Reference-only

See all 66 comparisons across every model →

Quick pill index

Direct comparisons against every other scored model on the bench:

Hy3 vs Fusion Hy3 vs Hermes MoA Hy3 vs Claude Fable 5 Hy3 vs Grok Hy3 vs MiniMax M3 Hy3 vs Fugu Ultra Hy3 vs GLM-5.2 Hy3 vs Fugu Mini Hy3 vs Opus 4.8 Hy3 vs Kimi K2.7 Hy3 vs Qwable 5 27B Coder Hy3 vs Claude Sonnet 5 Hy3 vs Qwen 3.7 Hy3 vs Agents-A1 Hy3 vs Gemma 4 12B · MLX Hy3 vs Laguna XS 2.1 Hy3 vs Qwythos 9B Hy3 vs LongCat-2.0 Hy3 vs Gemma-4 12B Coder

Read more on agentos.guide: /hy3-agent-os

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 3,600+ founders shipping with it every day all live inside the AI Profit Boardroom.

3,600+founders
258documented wins
38countries
$59/momonthly