Tencent Releases Hy3: An Open 295B Mixture-of-Experts (MoE) Model with 21B Active Parameters and 256K Context

Tencent’s Hy workforce launched Hy3. Hy3 is a 295B-parameter Mixture-of-Experts (MoE) mannequin. It prompts solely 21B parameters per token. The weights ship below the Apache License 2.0. Hy3 is geared toward reasoning, agentic workflows, and long-context duties.

What is Hy3?

Hy3’s structure incorporates a sparse MoE with 192 consultants and top-8 routing. Only 8 consultants hearth per token, so compute stays low.

The mannequin additionally makes use of a Multi-Token Prediction (MTP) layer. MTP predicts a number of tokens without delay for quicker decoding. Both vLLM and SGLang allow it via speculative decoding.

A separate Hy3-FP8 checkpoint can be launched. FP8 lowers the reminiscence footprint for cheaper serving.

Benchmark and Performance

The analysis workforce printed scores throughout coding, brokers, and STEM. On coding, Hy3 studies 78.0 on SWE-Bench Verified. It additionally studies 57.9 on SWE-Bench Pro and 75.8 on SWE-Bench Multilingual. Terminal-Bench 2.1 lands at 71.7, and DeepSWE at 28.0.

On STEM and reasoning, the numbers climb larger. Hy3 studies 90.4 on GPQA Diamond and 72.0 on USAMO 2026. IMOAnswerBench reaches 90.0, and HLE (with instruments) reaches 53.2.

The analysis workforce ran a blind take a look at with 270 consultants. That take a look at collected 312 legitimate comparisons on actual workflows. Hy3 scored 2.67 out of 4, forward of GLM-5.1 at 2.51. The edge was clearest in frontend growth, CI/CD, and knowledge and storage.

Reliability and Production Behavior

The analysis workforce targeted a lot of this launch on manufacturing reliability. Three failure modes acquired direct consideration, backed by inside numbers.

• Tool calling and output formatting: The workforce mounted baseline stability points that broke brokers. Invalid calls that set off infinite loops dropped. Hy3 additionally generalizes throughout agent scaffoldings. On SWE-Bench Verified, accuracy variance throughout CodeBuddy, Cline, and KiloCode stays inside 4%.
• World information and anti-hallucination: The goal conduct is straightforward: reply when grounded, flag when proof is lacking. In inside evaluations, the hallucination price fell from 12.5% to five.4%. Commonsense error charges fell from 25.4% to 12.7%.
• Multi-turn intent monitoring: Joint SFT and RL improved coreference and constraint monitoring. The inside difficulty price dropped from 17.4% to 7.9%. On the MRCR long-dialogue benchmark, scores rose from 42.9% to 75.1%.

How to Call Hy3

Hy3 exposes an OpenAI-compatible API. You deploy it with vLLM or SGLang, then name the endpoint. One flag, reasoning_effort, controls how a lot the mannequin thinks.

from openai import OpenAI

consumer = OpenAI(base_url="http://127.0.0.1:8000/v1", api_key="EMPTY")

response = consumer.chat.completions.create(
    mannequin="hy3",
    messages=[
        {"role": "user", "content": "Refactor this function and explain the change."},
    ],
    temperature=0.9,
    top_p=1.0,
    # reasoning_effort: "no_think" (default), "low", "excessive" (deep chain-of-thought)
    extra_body={"chat_template_kwargs": {"reasoning_effort": "excessive"}},
)
print(response.decisions[0].message.content material)

Use no_think for direct solutions, and excessive for math, coding, or multi-step duties. Tencent analysis workforce recommends temperature=0.9 and top_p=1.0. You may also attempt Hy3 with out native {hardware}. OpenRouter lists a tencent/hy3:free route at $0 per token. That free tier is scheduled to finish on July 21, 2026.

Where Hy3 Fits: Use Cases

Hy3 is constructed round agent-style, long-context work. Just a few concrete examples:

• Coding brokers: Feed a full repository into the 256K window. Ask Hy3 to repair a failing take a look at with reasoning_effort="excessive". Stable software calls assist it run edits throughout many information.
• Document processing: Pass a protracted contract or submitting as context. The anti-hallucination coaching reduces fabricated clauses and misquotes.
• Financial evaluation: Combine tables and prose in a single immediate. Ask for a grounded abstract that flags lacking knowledge moderately than guessing.
• Frontend and sport growth: Generate a React part or a small sport loop. The blind take a look at confirmed a frontend benefit over GLM-5.1.

Hy3 vs GLM-5.2

Tencent’s analysis workforce benchmarked Hy3 in opposition to GLM-5.2 in its personal appendix. GLM-5.2 is roughly a 744B MoE with about 40B lively parameters. Hy3 is lower than half that complete dimension, with 21B lively. On coding, GLM-5.2 leads throughout the suite.

The focus right here is about dimension, not simply rating. Hy3 trades some coding accuracy for a much smaller lively footprint. That footprint issues while you self-host and pay for GPUs.

Deployment Notes

Hy3 has 295B complete parameters, so serving wants actual reminiscence. Tencent’s analysis workforce recommends 8 GPUs, such because the H20-3e or playing cards with bigger reminiscence. vLLM and SGLang each ship recipes with MTP enabled. A minimal vLLM launch appears like this:

vllm serve tencent/Hy3 
  --tensor-parallel-size 8 
  --speculative-config.technique mtp 
  --speculative-config.num_speculative_tokens 2 
  --tool-call-parser hy_v3 
  --reasoning-parser hy_v3 
  --enable-auto-tool-choice 
  --port 8000 
  --served-model-name hy3

For compression, The analysis workforce factors to its AngelSlim toolkit. AngelSlim covers quantization, low-bit strategies, and speculative sampling. Tencent additionally offers an entire finetuning pipeline for Hy3.

Try It: Interactive Explorer

The demo under is an interactive explorer for Hy3. It visualizes MoE routing, reasoning modes, benchmarks, and sparse effectivity.