Hy3
- Input
- $0.14/M tok
- Output
- $0.58/M tok
- Blended
- $0.25/M tok · 3:1
- Speed
- 86tok/s
- Context
- 256Ktokens
- Cost per task
- ~$0.24/taskest.
Open weights.
Rankings
Coding
Mixed#81 / 137rank by quality39.2±4.3quality / 1005/7 benchmarks measuredWriting & chat
Verified#43 / 162rank by quality64.2±6.0quality / 1002/5 benchmarks measuredHard reasoning
Verified#67 / 218rank by quality60.2±3.7quality / 1003/4 benchmarks measuredAgents
Mixed#59 / 187rank by quality44.5±4.6quality / 1004/7 benchmarks measured
Self-hostingEstimated
How we estimate →What it would cost to run these open weights yourself on rented GPUs, compared with the API price above.
- Hardware
- 4× B200 180GB
- FP8 weights
- Throughput
- ~16,423 tok/s
- many requests batched
- Price
- $0.17–1.50 /M tok
- busy → light use
- On your own machine
- Mac Studio M5 Ultra 512GB (int4, ~80 tok/s single-stream)
- single consumer GPU or Mac
Assumptions (5)
- FP8 weights (295B params, 21B active per token) + 25% KV-cache headroom ≈ 369 GB
- 4× B200 180GB at $5.98–$14.24/GPU-hour on-demand (2026-09-24)
- ~16,423 output tok/s aggregate at batch 470 (bandwidth-bound); MoE compute scales with active params
- Blended 3:1 input:output; prefill ~202,500 tok/s
- Low = 75% utilization at the low GPU price; high = 20% at the high price
Benchmark scores
~ italic, dashed = no published score yet, estimated from related benchmarks.
Coding
DeepSWE v1.1Lab-reported
#43 of 4528.0%
- #1Gemini 4 Argon78.8%
- #2DeepSeek V4.1 Flash 101777.0%
- #42Nex-N2.5-Mini36.1%
- #43Hy328.0%
- #44Nemotron 3.5 Super VL24.6%
huggingface.co · 2026-07-06
LiveCodeBenchEstimated
~#15 of 28 (estimated)~73.4%
- #1Qwen3.8-Omni-Flash92.6%
- #2Qwen3.8 Flash-Next91.9%
- #14Maple-Preview75.1%
- ~#15Hy3Estimated73.4%
- #15Mistral Small 464.0%
No published score — estimated from related benchmarks
SciCodeVerified
#47 of 11848.6%
- #1Claude Fable 5.267.7%
- #2Claude Opus 5.566.9%
- #46Beam 1.0 501B A23B49.2%
- #47Hy348.6%
- #48Qwen3.8 27B NVFP448.4%
artificialanalysis.ai · 2026-09-24
AA Coding Agent IndexEstimated
~#24 of 26 (estimated)~33.3
- #1Claude Opus 5.561.7
- #2Claude Opus 559.7
- #23Gemini 3.8 Flash33.3
- ~#24Hy3Estimated33.3
- #24GPT-6 Luna31.1
No published score — estimated from related benchmarks
Writing & chat
IFBenchEstimated
~#52 of 140 (estimated)~64.8%
- #1MiniMax-M382.9%
- #2Qwen3.8 2.4T A95B82.8%
- #51gpt-oss-20b65.1%
- ~#52Hy3Estimated64.8%
- #52Qwen3.6 35B A3B64.4%
No published score — estimated from related benchmarks
EQ-Bench Creative Writing v3 (Elo)Estimated
~#13 of 26 (estimated)~1831 Elo
- #1GPT-6 Astra2173
- #2Claude Fable 5.12162
- #12Qwen3.8 2.4T A95B1843
- ~#13Hy3Estimated1831
- #13Muse Glimmer1798
No published score — estimated from related benchmarks
Hard reasoning
GPQA DiamondVerified
#33 of 19289.7%
- #1GPT-6 Astra96.1%
- #2Gemini 3.8 Flash95.3%
- #32Nemotron 3.5 Super VL NVFP489.9%
- #33Hy389.7%
- #34Kimi K2.7 Code89.6%
artificialanalysis.ai · 2026-09-24
AIME (latest)Estimated
~#9 of 34 (estimated)~92.6%
- #1GLM-5.3 NVFP499.5%
- #2Beam 1.0 501B A23B97.7%
- #8gpt-oss-120b93.4%
- ~#9Hy3Estimated92.6%
- #9Qwen3.8 27B MXFP492.5%
No published score — estimated from related benchmarks
Agents
τ²-bench (Telecom)Estimated
~#38 of 120 (estimated)~76.1%
- #1Step 3.7 Flash98.5%
- #2DeepSeek-V4-Pro-0813-nvfp4-DSpark98.3%
- #37EXAONE 4.5 33B78.1%
- ~#38Hy3Estimated76.1%
- #38GPT-5.4 nano76.0%
No published score — estimated from related benchmarks
OSWorld-VerifiedEstimated
~#11 of 15 (estimated)~69.1%
No published score — estimated from related benchmarks
OSWorld 2.0Estimated
~last of 5 (estimated)below measured range (<47.5%)
No published score — estimated from related benchmarks
GDPval-AA (Elo)Verified
#39 of 911045 Elo
- #1Claude Opus 5.51866
- #2Muse Wonder Full1857
- #38Grok Build 0.1 06161052
- #39Hy31045
- #40Kimi K2.7 Code1025
artificialanalysis.ai · 2026-09-24