ModelBench
Tencent logo
Models/Tencent

Hy3 preview

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

Source-linked Hy Open weights Released 2026-04-20
API data Report
Input modalitiesT
Output modalitiesT
Input price$0.063/1M tokens
Output price$0.21/1M tokens
Context window256K
Max output64K
Overview

Capabilities and specifications

✓ Reasoning Yes
✓ Tool calling Yes
? Structured output Unknown
× Attachments No
× Vision input No
✓ Open weights Yes
Creator
Tencent
Model family
Hy
Knowledge cutoff
Not documented
License
Not documented
Release date
2026-04-20
Model ID
tencent/hy3-preview
Inference availability

Provider pricing and availability

Provider-specific identifiers, limits, and prices per million tokens.

Report a price
Providers offering Hy3 preview
ProviderProvider model IDContextInputOutputCache readCapabilitiesSource
Tencent TokenHub hy3-preview 256K Not listed Not listed - ReasoningTools Docs
OpenRouter tencent/hy3-preview 262.144K $0.063 $0.21 $0.021 ReasoningTools Docs
SiliconFlow tencent/Hy3-preview 262.144K $0.066 $0.26 $0.029 ReasoningTools Docs
Kilo Gateway tencent/hy3-preview 262.144K $0.066 $0.26 $0.029 ReasoningTools Docs
NanoGPT tencent/hy3-preview 262.144K $0.066 $0.26 $0.029 Not listed Docs
ZenMux tencent/hy3-preview 256K $0.172 $0.572 $0.058 ReasoningTools Docs
CrossModel tencent/hy3-preview 262.144K $0.19 $0.63 $0.063 ReasoningTools Docs

Capability badges appear only when the provider catalog explicitly lists support.

Serving quality

Performance and reliability

Observed serving measurements stay separate from model specifications and are never estimated from price or model size.

Throughput Not reported

Generation speed; higher is faster.

Latency / TTFT Not reported

Time until the first output token arrives.

Uptime Not reported

Successful requests over a measured period.

Performance varies by inference provider, region, and load. Provider-level measurements appear only when a source reports them.

Published evaluations

Benchmarks

Raw results remain attached to their original source and version.

Benchmark registry
Agentic Index - Unversioned - indexTop 8 reported models
Claude Opus 5
55.3
GPT-5.6 Sol
54.0
Anthropic: Claude Fable 5 (batch)
52.8
Claude Fable 5
52.8
Kimi K3
50.1
GPT-5.6 Terra
47.4
Anthropic: Claude Opus 4.8 (batch)
47.2
Hy3 preview
30.7
Coding Index - Unversioned - indexTop 8 reported models
Claude Opus 5
78.0
GPT-5.6 Sol
77.4
GPT-5.6 Terra
76.7
Anthropic: Claude Fable 5 (batch)
76.5
Claude Fable 5
76.5
Kimi K3
76.2
OpenAI: GPT-5.5 (batch)
74.9
Hy3 preview
58.8
Intelligence Index - Unversioned - indexTop 8 reported models
Claude Opus 5
60.7
Anthropic: Claude Fable 5 (batch)
59.9
Claude Fable 5
59.9
GPT-5.6 Sol
58.9
Kimi K3
57.1
Anthropic: Claude Opus 4.8 (batch)
55.7
Anthropic: Claude Opus 4.8
55.7
Hy3 preview
41.2
SWE-Bench Verified - Unversioned - resolvedTop 8 reported models
Claude Fable 5
95.0
Claude Opus 4.8
88.6
Claude Sonnet 5
85.2
DeepSeek V4 Pro
80.6
Qwen3.7 Max
80.4
Kimi K2.6
80.2
DeepSeek V4 Flash
79.0
Hy3 preview
74.4
Catalog activity

Recent record changes

Detected changes from successful source imports.

Full change log
Context Length262144 → 256000
Context Length256000 → 262144
Context Length262144 → 256000
Context Length256000 → 262144
Context Length262144 → 256000
Context Length256000 → 262144
Context Length262144 → 256000
Context Length256000 → 262144
Public API

Use this record

Fetch the complete source-linked model record without an API key.

API documentation
Endpoint
GET https://www.modelbench.lol/api/v1/models/tencent/hy3-preview
curl
curl "https://www.modelbench.lol/api/v1/models/tencent/hy3-preview"