Hy3

Tool callingStructured outputReasoning

Hugging Face Launch post Website

Tencent Hy reasoning model for coding, instruction following, and agent tasks

Context: 262K
Released: 3mo ago
Parameters: —
Input / 1M tokens: $0.08
Output / 1M tokens: $0.33
Cached input / 1M tokens: $0.02

Sources:models.devOpenRouter

Benchmarks

DeepSWE is the primary coding benchmark where an exact model evaluation exists. Other sourced evaluations appear below; missing scores are not zero.

Not evaluated in DeepSWE 1.1 as of 2026-09-22. Results are only shown for exact model versions in the official dataset. Official leaderboard.

No benchmark results have been matched to this model yet.

Specifications

Developer: Tencent
Model family: Hunyuan
Type: Language model
Predecessor: —
Successor: —
Released: July 6, 2026
Model ID: tencent/hy3
Openness: Open weights
License: —
Parameters: —
Input modalities: text
Output modalities: text
Modality: text->text
Tokenizer: Other
Max output tokens: 128K
Hugging Face: tencent/Hy3
Source freshness: Metadata synced Oct 8, 2026, 10:17 PM UTC

Additional source evidence

Source model: tencent/hy3
Source family: Hy
Last updated: 2026-07-06
Model context: 256K
Maximum input: 192K
Model output limit: 128K
Open weights: Yes
File attachments: No
Reasoning: Yes
Tool calling: Yes
Temperature control: Yes

Downloadable weights

Hugging Face

Published benchmarks

Benchmark	Score	Evaluation details	Evidence
SWE-Bench Verified	78 · resolved	—	Source
SWE-Bench Multilingual	75.8 · score	SWE-agent · highest reasoning effort	Source
SWE-Bench Pro	57.9 · score	SWE-agent · highest reasoning effort	Source
Terminal-Bench	71.7 · score	Terminus 2 · highest reasoning effort; 4h timeout; 500 episodes · 2.1	Source
NL2Repo	45.6 · score	Claude Code · highest reasoning effort; 250 turns; 12000s timeout	Source
DeepSWE	28 · score	mini-swe-agent · highest reasoning effort; 2h timeout	Source
BrowseComp	84.2 · score	Tencent internal search harness · highest reasoning effort	Source
WideSearch	76.4 · score	Tencent internal search harness · highest reasoning effort	Source
DeepSearchQA	91 · score	Tencent internal search harness · highest reasoning effort	Source
MCP Atlas	79.1 · score	Scale AI · 500 public tasks · highest reasoning effort; April 2026; 100 tool calls	Source
Toolathlon	48.5 · score	highest reasoning effort	Source
APEX-Agents	25.6 · pass@1	highest reasoning effort	Source
ClawEval	68.5 · pass@3	Tencent internal harness · 105 queries · highest reasoning effort · 20260325	Source
WildClawBench	53.6 · score	OpenClaw · 35 text-only tasks · highest reasoning effort	Source
SkillsBench	55.3 · average over 3 runs	Claude Code · 79 text-only tasks · highest reasoning effort	Source
Humanity's Last Exam	53.2 · score	text-only · highest reasoning effort; with tools	Source
Humanity's Last Exam	37 · score	text-only · highest reasoning effort; without tools	Source
GPQA Diamond	90.4 · score	highest reasoning effort	Source
FrontierScience	21.3 · score	highest reasoning effort; research	Source
FrontierScience	74.8 · score	highest reasoning effort; olympiad	Source
USAMO	72 · score	highest reasoning effort · 2026	Source
MathArena Apex	38.7 · score	highest reasoning effort	Source
ArxivMath	52.2 · score	highest reasoning effort	Source
HorizonMath	7.1 · pass@12	highest reasoning effort	Source
PHYBench	77.4 · score	highest reasoning effort	Source
CMT Benchmark	37.8 · score	highest reasoning effort	Source
IMOAnswerBench	90 · score	highest reasoning effort	Source
SuperChem	54.9 · score	highest reasoning effort	Source
CL-bench	23.8 · score	highest reasoning effort	Source
CL-bench-life	17 · score	highest reasoning effort	Source
AA-LCR	73.4 · score	highest reasoning effort	Source

Complete source record (JSON)

{
  "id": "tencent/hy3",
  "name": "Hy3",
  "description": "Tencent Hy reasoning model for coding, instruction following, and agent tasks",
  "family": "Hy",
  "last_updated": "2026-07-06",
  "attachment": false,
  "reasoning": true,
  "tool_call": true,
  "temperature": true,
  "modalities": {
    "input": [
      "text"
    ],
    "output": [
      "text"
    ]
  },
  "limit": {
    "context": 256000,
    "input": 192000,
    "output": 128000
  },
  "benchmarks": [
    {
      "name": "SWE-Bench Verified",
      "score": 78,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "resolved"
    },
    {
      "name": "SWE-Bench Multilingual",
      "score": 75.8,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "harness": "SWE-agent",
      "variant": "highest reasoning effort"
    },
    {
      "name": "SWE-Bench Pro",
      "score": 57.9,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "harness": "SWE-agent",
      "variant": "highest reasoning effort"
    },
    {
      "name": "Terminal-Bench",
      "score": 71.7,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "harness": "Terminus 2",
      "variant": "highest reasoning effort; 4h timeout; 500 episodes",
      "version": "2.1"
    },
    {
      "name": "NL2Repo",
      "score": 45.6,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "harness": "Claude Code",
      "variant": "highest reasoning effort; 250 turns; 12000s timeout"
    },
    {
      "name": "DeepSWE",
      "score": 28,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "harness": "mini-swe-agent",
      "variant": "highest reasoning effort; 2h timeout"
    },
    {
      "name": "BrowseComp",
      "score": 84.2,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "harness": "Tencent internal search harness",
      "variant": "highest reasoning effort"
    },
    {
      "name": "WideSearch",
      "score": 76.4,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "harness": "Tencent internal search harness",
      "variant": "highest reasoning effort"
    },
    {
      "name": "DeepSearchQA",
      "score": 91,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "harness": "Tencent internal search harness",
      "variant": "highest reasoning effort"
    },
    {
      "name": "MCP Atlas",
      "score": 79.1,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "harness": "Scale AI",
      "variant": "highest reasoning effort; April 2026; 100 tool calls",
      "dataset": "500 public tasks"
    },
    {
      "name": "Toolathlon",
      "score": 48.5,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort"
    },
    {
      "name": "APEX-Agents",
      "score": 25.6,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "pass@1",
      "variant": "highest reasoning effort"
    },
    {
      "name": "ClawEval",
      "score": 68.5,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "pass@3",
      "harness": "Tencent internal harness",
      "variant": "highest reasoning effort",
      "dataset": "105 queries",
      "version": "20260325"
    },
    {
      "name": "WildClawBench",
      "score": 53.6,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "harness": "OpenClaw",
      "variant": "highest reasoning effort",
      "dataset": "35 text-only tasks"
    },
    {
      "name": "SkillsBench",
      "score": 55.3,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "average over 3 runs",
      "harness": "Claude Code",
      "variant": "highest reasoning effort",
      "dataset": "79 text-only tasks"
    },
    {
      "name": "Humanity's Last Exam",
      "score": 53.2,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort; with tools",
      "dataset": "text-only"
    },
    {
      "name": "Humanity's Last Exam",
      "score": 37,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort; without tools",
      "dataset": "text-only"
    },
    {
      "name": "GPQA Diamond",
      "score": 90.4,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort"
    },
    {
      "name": "FrontierScience",
      "score": 21.3,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort; research"
    },
    {
      "name": "FrontierScience",
      "score": 74.8,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort; olympiad"
    },
    {
      "name": "USAMO",
      "score": 72,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort",
      "version": "2026"
    },
    {
      "name": "MathArena Apex",
      "score": 38.7,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort"
    },
    {
      "name": "ArxivMath",
      "score": 52.2,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort"
    },
    {
      "name": "HorizonMath",
      "score": 7.1,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "pass@12",
      "variant": "highest reasoning effort"
    },
    {
      "name": "PHYBench",
      "score": 77.4,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort"
    },
    {
      "name": "CMT Benchmark",
      "score": 37.8,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort"
    },
    {
      "name": "IMOAnswerBench",
      "score": 90,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort"
    },
    {
      "name": "SuperChem",
      "score": 54.9,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort"
    },
    {
      "name": "CL-bench",
      "score": 23.8,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort"
    },
    {
      "name": "CL-bench-life",
      "score": 17,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort"
    },
    {
      "name": "AA-LCR",
      "score": 73.4,
      "source": "https://huggingface.co/tencent/Hy3",
      "metric": "score",
      "variant": "highest reasoning effort"
    }
  ],
  "release_date": "2026-07-06",
  "open_weights": true,
  "weights": [
    {
      "label": "Hugging Face",
      "url": "https://huggingface.co/tencent/Hy3"
    }
  ]
}

Infrastructure

Provider availability

All providers

Provider	Endpoint	Context	Input /M tokens	Output /M tokens	Cached input /M tokens	Cache write /M tokens	Uptime 1d	Data handling
Tencent	tencent/fp8	262K	$0.13	$0.53	$0.03	—	100.00%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
GMICloud	gmicloud/bf16	262K	$0.14	$0.58	$0.04	—	100.00%	ZDR: No Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
Novita	novita	262K	$0.14	$0.58	$0.04	—	100.00%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
Phala	phala	262K	$0.15	$0.64	$0.04	—	99.78%	ZDR: Yes Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15
AtlasCloud	atlas-cloud/fp8	262K	$0.20	$0.80	$0.05	—	100.00%	ZDR: No Training: Unknown Prompt logging: Unknown Retention: Unknown ZDR source Privacy policy Checked 2026-10-15

Provider data:OpenRouter

Deployment graph

Can I run this?

Vertical Tabs only shows compatibility when it has structured evidence tying this exact model to hardware, runtime, quantization, memory use, and observed performance.

Browse hardware