Hy3
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
- $0.0825 / $0.33 per 1M
- Context
- 262K tokens
- Released
- Jul 6, 2026
Paste into Claude Code, Cursor, or Codex
Help me set up the Respan gateway: fetch and follow https:/
Providers
3 providers, cheapest first, OpenRouter last, priced per 1M tokens after gateway discounts.
| Provider | Input | Output |
|---|---|---|
Concentrate concentrate/hy3 | $0.13/M | $0.53/M |
Novita novita/hy3 | $0.14/M | $0.58/M |
OpenRouter openrouter/tencent/hy3 | $0.0825/M | $0.33/M |
Is a provider down? OpenRouter status
Pricing
Averaged across 3 providers, in USD per 1M tokens after gateway discounts.
- Average input
- $0.12 per 1M
- Average output
- $0.48 per 1M
- Average cache read
- $0.03 per 1M
Cost by volume
Total cost from 0 to 1T tokens on the cheapest and most expensive provider.
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 better5.3tokens/sacross providers
Response latency
Lower is better1.8sacross providers
Benchmarks
The Respan Index plus 44 published scores from 3 sources, each linked to where it was published.
39.5/ 100
Respan Index#71 of 176
- 71Hy339.5
- 69Gemini 3 Pro Preview39.7
- 71GPT-539.5
Published scores
| Benchmark | Score | Source | Published |
|---|---|---|---|
| IMOAnswerBenchgithub.com | 90% | github.com | Jul 6, 2026 |
| SWE-bench Verifiedgithub.com | 78% | github.com | Jul 6, 2026 |
| SWE-bench Multilingualgithub.com | 75.8% | github.com | Jul 6, 2026 |
| Terminal-Bench 2.1github.com | 71.7% | github.com | Jul 6, 2026 |
| SWE-bench Progithub.com | 57.9% | github.com | Jul 6, 2026 |
| Hy-SWE Maxgithub.com | 49% | github.com | Jul 6, 2026 |
| NL2Repogithub.com | 45.6% | github.com | Jul 6, 2026 |
| DeepSWEgithub.com | 28% | github.com | Jul 6, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| DeepSearchQAgithub.com | 91% | github.com | Jul 6, 2026 |
| BrowseCompgithub.com | 84.2% | github.com | Jul 6, 2026 |
| MCP Atlasgithub.com | 79.1% | github.com | Jul 6, 2026 |
| WideSearchgithub.com | 76.4% | github.com | Jul 6, 2026 |
| ClawEvalgithub.com | 68.5% | github.com | Jul 6, 2026 |
| Toolathlon-VerifiedHKUST | 56.8% | HKUST | Jul 30, 2026 |
| SkillsBenchgithub.com | 55.3% | github.com | Jul 6, 2026 |
| WildClawBenchgithub.com | 53.6% | github.com | Jul 6, 2026 |
| Toolathlongithub.com | 48.5% | github.com | Jul 6, 2026 |
| Apex-Agentgithub.com | 25.6% | github.com | Jul 6, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| GPQA Diamondgithub.com | 90.4% | github.com | Jul 6, 2026 |
| PHYBenchgithub.com | 77.4% | github.com | Jul 6, 2026 |
| FrontierScience-Olympiadgithub.com | 74.8% | github.com | Jul 6, 2026 |
| AA-LCRgithub.com | 73.4% | github.com | Jul 6, 2026 |
| SuperChemgithub.com | 54.9% | github.com | Jul 6, 2026 |
| HLE (with tools, text-only)github.com | 53.2% | github.com | Jul 6, 2026 |
| Humanity's Last Exam (no tools)github.com | 37% | github.com | Jul 6, 2026 |
| FrontierScience-Researchgithub.com | 21.3% | github.com | Jul 6, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| USAMO 2026github.com | 72% | github.com | Jul 6, 2026 |
| Hy-Mathgithub.com | 60.9% | github.com | Jul 6, 2026 |
| MathArena Apexgithub.com | 38.7% | github.com | Jul 6, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| CL-benchgithub.com | 23.8% | github.com | Jul 6, 2026 |
| CL-bench Lifegithub.com | 17% | github.com | Jul 6, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| LMArena WebDevLMArena | 1506 Elo · #64 of 124 | LMArena | Nov 19, 2026 |
| LMArena EloLMArena | 1454 Elo · #62 of 191 | LMArena | Nov 19, 2026 |
| LMArena AgentLMArena | -0.0863 · #57 of 68 | LMArena | Nov 18, 2026 |
| Benchmark | Score | Source | Published |
|---|---|---|---|
| Hy-FinModelBenchgithub.com | 69% | github.com | Jul 6, 2026 |
| ArxivMathgithub.com | 52.2% | github.com | Jul 6, 2026 |
| e-benchgithub.com | 50.2% | github.com | Jul 6, 2026 |
| Hy-SkillsWorldgithub.com | 45.8% | github.com | Jul 6, 2026 |
| Hy-CompanyBenchgithub.com | 41.7% | github.com | Jul 6, 2026 |
| CMT-Benchmarkgithub.com | 37.8% | github.com | Jul 6, 2026 |
| Hy-Backend 2.0github.com | 25% | github.com | Jul 6, 2026 |
| Hy-Euler Progithub.com | 24.2% | github.com | Jul 6, 2026 |
| ProdBenchgithub.com | 23% | github.com | Jul 6, 2026 |
| HorizonMathgithub.com | 7.1% | github.com | Jul 6, 2026 |
Compare Hy3: Hy3 vs Hy4, Hy3 vs DeepSeek V4.1 Flash, Hy3 vs Hy4 preview, Hy3 vs Hy3 preview
Apps
Use Hy3 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 FAQ
Short answers, built from the data on this page.
Hy3 is a model from Tencent, released on Jul 6, 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 costs from $0.0825 per 1M input tokens and $0.33 per 1M output tokens on OpenRouter, after gateway discounts. Averaged across its 3 providers, it costs $0.12 per 1M input tokens and $0.48 per 1M output tokens.
Hy3 is available through 3 providers: Concentrate, Novita, and OpenRouter.
Hy3 accepts up to 262K tokens of context on its largest provider route.
Point any OpenAI-compatible SDK at https://gateway.respan.ai/v1 with your Respan API key and set the model to "hy3". To set it up with a coding agent, copy the agent setup prompt at the top of this page.
Hy3 scores 39.5 out of 100 on the Respan Index v1.1, ranking #71 of 176 models. By category: Coding 16.4, Agents 51.1, Reasoning 43.4, Knowledge & language 37.5 (estimated), and Human preference 53.7. The Respan Index combines independent and lab-reported benchmark results into one score, using benchmark data collected by Respan.
Hy3 has 44 published benchmark scores, including IMOAnswerBench 90%, DeepSearchQA 91%, and GPQA Diamond 90.4%. 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.