Skip to content
llmwaves

Hy3

Tencent · released Jul 6, 2026

Open weightsReasoningTool useStructured output

Hy3 is a 295B-parameter Mixture-of-Experts model from Tencent (21B active, 192 experts with top-8 routing) built for reasoning, agentic workflows, and real-world production use. It supports a configurable reasoning effort:...

Specification

Context window
262K
Max output
128K
Knowledge cutoff
Not stated
Parameters
298.8B
Licence
apache-2.0
Serving providers
6
Moderated
No
Uptime
99.9%

Intelligence

42.2

91th percentile

Coding

58.8

Coding Index

Agentic

31.4

Agentic Index

Output speed

78 t/s

Median across providers

Latency

434ms

Time to first token

Cost per task

$0.05

Estimated

Benchmarks

Where the score comes from

The Intelligence Index is a composite. These are the underlying evaluations this model was actually measured on.

Evaluation scores

Percentage correct · higher is better

  • GPQA Diamond
    89.7%
  • AA-LCR (long context)
    74.7%
  • SciCode
    47.6%
  • Humanity's Last Exam
    33.5%
  • τ²-bench (Telecom)
    22.9%

An evaluation missing from this list was not run for this model — it is not a zero.

View as table
EvaluationScore
GPQA Diamond89.7%
AA-LCR (long context)74.7%
SciCode47.6%
Humanity's Last Exam33.5%
τ²-bench (Telecom)22.9%

Against its peers

Intelligence Index · this model highlighted, nearest peers in grey

Peers are the models sitting closest on the Intelligence Index — the set you would realistically choose between.

View as table
ModelIntelligence
GPT-5.243.3
Kimi K2.7 Code43.0
MiMo-V2.5-Pro42.9
Inkling42.3
Hy342.2
Nex-N2-Pro41.7
Inkling Small41.2
GPT-5.2-Codex41.2

Percentile among all indexed models

Intelligence91th
Coding76th
Agentic61th
Arena Elo67th

Pricing

What it costs to run

List prices per million tokens, plus what one representative task works out to.

List price

Input / 1M tokens
$0.132
Output / 1M tokens
$0.528
Cached input / 1M
$0.033
Blended 3:1
$0.231

One task, estimated

$0.05

Input tokens
50,000
Output tokens
80,000
Profile
Reasoning

Estimated from list pricing: 50K input tokens plus 80K output tokens for reasoning models (25K for non-reasoning).

Serving providers

Speed and latency figures are medians across these providers, so a widely-served model reports a blend rather than any single endpoint.

Arena

Head-to-head generation quality

Elo from pairwise judgements, broken out by the kind of thing the model was asked to build.

Overall Elo
1,231
Win rate
43.8%
Strongest at
3D scenes
Tournaments
754

Elo by category

Dot position on a 1,150–1,250 scale · Elo has no meaningful zero

Agent categories (full-stack apps, mobile apps) are only scored for models tested in agent harnesses.

View as table
CategoryEloWin rate
3D scenes123143.8%
Websites120541.4%
UI components119840.2%
Game development117938.6%
Data visualisation115436.1%