Hy3 Preview
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: a direct no-think mode by default, plus low and high chain-of-thought modes for complex math, coding, and multi-step problems. With a 256K context window, Hy3 targets long-horizon tasks, including improved coreference resolution, multi-turn constraint tracking, and stable tool-calling that generalizes across agent scaffoldings. Tencent positions it as a reliable, cost-effective option across coding, document processing, financial analysis, game development, and frontend design, with a strong emphasis on grounded, anti-hallucination behavior that answers when grounded and flags when evidence is missing rather than fabricating.
- Input / output price
- $0.18 / $0.60 per 1M
- Context
- 262K tokens
- Released
- Apr 22, 2026
Paste into Claude Code, Cursor, or Codex
Help me set up the Respan gateway: fetch and follow https:/
Providers
One provider serves this model, priced per 1M tokens after gateway discounts.
| Provider | Input | Output |
|---|---|---|
OpenRouter openrouter/hy3-preview | $0.18/M | $0.60/M |
Is a provider down? OpenRouter status
Pricing
USD per 1M tokens on OpenRouter, after gateway discounts.
- Input
- $0.18 per 1M
- Output
- $0.60 per 1M
- Cache read
- $0.06 per 1M
Cost by volume
Total cost from 0 to 1T tokens.
Throughput and latency
30-day averages from requests through the Respan gateway. Under each figure, every model on the gateway by its average, this one in blue.
Output speed
Higher is better3.2tokens/sacross providers
Response latency
Lower is better2.1sacross providers
Benchmarks
The Respan Index plus 17 published scores from 2 sources, each linked to where it was published.
28.3/ 100
Respan Index#97 of 175
- 97Hy3 preview28.3
- 96NVIDIA Nemotron 3 Ultra 550B-A55B NVFP428.4
- 98GLM-527.8
Published scores
| Benchmark | Score | Source | Published |
|---|---|---|---|
| IMO Answer Benchgithub.com | 84.3% | github.com | Apr 23, 2026 |
| SWE-bench Verifiedgithub.com | 74.4% | github.com | Apr 23, 2026 |
| Terminal-Bench 2.0github.com | 54.4% | github.com | Apr 23, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| WideSearchgithub.com | 70.2% | github.com | Apr 23, 2026 |
| BrowseCompgithub.com | 67.1% | github.com | Apr 23, 2026 |
| ClawEval (pass^3)github.com | 55% | github.com | Apr 23, 2026 |
| WildClawBench (text-only)github.com | 45.3% | github.com | Apr 23, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| GPQA-Diamondgithub.com | 87.2% | github.com | Apr 23, 2026 |
| FrontierScience Olympiadgithub.com | 70% | github.com | Apr 23, 2026 |
| LongBench v2github.com | 65.4% | github.com | Apr 23, 2026 |
| HLE (text-only)github.com | 30% | github.com | Apr 23, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| CL-benchgithub.com | 22.8% | github.com | Apr 23, 2026 |
| CL-bench Lifegithub.com | 15.7% | github.com | Apr 23, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| LMArena EloLMArena | 1409 Elo · #114 of 187 | LMArena | Nov 18, 2026 |
| LMArena WebDevLMArena | 1359 Elo · #99 of 123 | LMArena | Nov 18, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| CHSBO 2025 (text-only, pass^3)github.com | 87.8% | github.com | Apr 23, 2026 |
| AdvancedIF (CIF&CC subsets)github.com | 79.5% | github.com | Apr 23, 2026 |
Compare Hy3 Preview: Hy3 preview vs Hy4 preview, Hy3 preview vs Hy3
Apps
Use Hy3 Preview in the tools you already run. Each guide points the app at the Respan gateway.
Setup guides for more apps are in the integration docs.
Hy3 Preview FAQ
Short answers, built from the data on this page.
Hy3 Preview is a model from Tencent, released on Apr 22, 2026. 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: a direct no-think mode by default, plus low and high chain-of-thought modes for complex math, coding, and multi-step problems. With a 256K context window, Hy3 targets long-horizon tasks, including improved coreference resolution, multi-turn constraint tracking, and stable tool-calling that generalizes across agent scaffoldings. Tencent positions it as a reliable, cost-effective option across coding, document processing, financial analysis, game development, and frontend design, with a strong emphasis on grounded, anti-hallucination behavior that answers when grounded and flags when evidence is missing rather than fabricating.
On the Respan gateway, Hy3 Preview costs $0.18 per 1M input tokens and $0.60 per 1M output tokens on OpenRouter, after gateway discounts.
Hy3 Preview is served by OpenRouter.
Hy3 Preview accepts up to 262K tokens of context.
Point any OpenAI-compatible SDK at https://gateway.respan.ai/v1 with your Respan API key and set the model to "hy3-preview". To set it up with a coding agent, copy the agent setup prompt at the top of this page.
Hy3 Preview scores 28.3 out of 100 on the Respan Index v1.1, ranking #97 of 175 models. By category: Coding 11.5, Agents 35.6, Reasoning 27.0 (estimated), Knowledge & language 23.2 (estimated), and Human preference 49.8. The Respan Index combines independent and lab-reported benchmark results into one score, using benchmark data collected by Respan.
Hy3 Preview has 17 published benchmark scores, including IMO Answer Bench 84.3%, WideSearch 70.2%, and GPQA-Diamond 87.2%. Each score links to where it was published; the data is collected by Respan.
Certified where it counts.
Built to meet the security and privacy standards that enterprise and healthcare teams require.
ISO 27001
The internationally recognized standard for information security management.
SOC 2
Secure, compliant management of your data across all of our systems.
GDPR
Operated under GDPR, the world's strictest standard for data privacy.
HIPAA
HIPAA compliant, with a BAA available for healthcare teams.