{"slug": "tencent-hy3-295b-moe-hits-swe-bench-78-free-api-ends-july-21", "title": "Tencent Hy3: 295B MoE Hits SWE-Bench 78 — Free API Ends July 21", "summary": "Tencent released Hy3, a 295B Mixture-of-Experts model with Apache 2.0 weights, achieving a 78.0 SWE-bench Verified score and leading open-weight models on tool and search benchmarks. A free API on OpenRouter is available until July 21, after which pricing will be $0.20 per million input tokens and $0.80 per million output tokens, offering a cost-effective alternative to proprietary models like Claude Sonnet 5.", "body_md": "Tencent dropped Hy3 on July 6 — a 295B Mixture-of-Experts model with Apache 2.0 weights, a 78.0 SWE-bench Verified score, and a free API on OpenRouter that expires July 21. If you’re choosing a coding agent backend or evaluating open-weight models for agentic workflows, you have 13 days to test it at zero cost.\n\n## What the SWE-Bench Score Actually Means\n\nHy3 hits 78.0 on SWE-bench Verified. That number deserves context before you act on it.\n\n[GLM-5.2](https://byteiota.com/zcode-glm-52-china-data-risk/) scores 84.2 on the same benchmark. Claude Sonnet 5 leads the full BenchLM leaderboard at around 87 across agentic, coding, and reasoning tasks. So Hy3 is not the top coder — and Tencent doesn’t pretend it is.\n\nWhere Hy3 leads is tool and search workloads: 84.2 on BrowseComp, 91.0 on DeepSearchQA, and 79.1 on MCP-Atlas tool orchestration — best in the open-weight field on all three. If your agent crawls the web, calls MCP servers, or does long-context retrieval, Hy3’s benchmark profile fits that workload better than its SWE-bench number suggests.\n\nBenchmark stability is also worth noting. Tencent tested Hy3 across KiloCode, Cline, and CodeBuddy scaffoldings and reported less than 4% accuracy variance. That kind of consistency across frameworks matters in production — a model that performs differently depending on your agent harness is a reliability problem.\n\n## The Free API Closes July 21\n\nOpenRouter’s `tencent/hy3:free`\n\nendpoint is live now. Cost: $0 per million tokens, input or output. On July 21, that flips to $0.20 per million input tokens and $0.80 per million output tokens.\n\nAfter the free window closes, Hy3 is still substantially cheaper than alternatives — [Claude Sonnet 5](https://byteiota.com/claude-sonnet-5-migration-three-breaking-api-changes/) runs around $3/$15 per million tokens, making Hy3 roughly 15x cheaper at the post-July-21 rate. But you won’t get another free evaluation window. The time to test it is now.\n\nThe API is OpenAI-compatible. Swap your base URL, change the model name, done:\n\n``` python\nfrom openai import OpenAI\nimport os\n\nclient = OpenAI(\n    api_key=os.getenv(\"OPENROUTER_API_KEY\"),\n    base_url=\"https://openrouter.ai/api/v1\"\n)\n\nresponse = client.chat.completions.create(\n    model=\"tencent/hy3:free\",\n    messages=[{\"role\": \"user\", \"content\": \"Review this function for bugs: ...\"}],\n    extra_body={\"reasoning_effort\": \"high\"}\n)\nprint(response.choices[0].message.content)\n```\n\nThe `reasoning_effort`\n\nparameter controls chain-of-thought depth: `\"no_think\"`\n\nfor fast direct responses, `\"low\"`\n\nfor light reasoning, `\"high\"`\n\nfor complex tasks like multi-step code review or debugging. Set it wrong and you’re either wasting tokens or missing accuracy.\n\n## Apache 2.0 — No Regional Restrictions\n\nThe April 2026 preview version excluded the EU, UK, and South Korea from its license. The July 6 official release fixes that. Hy3 is now Apache 2.0 with zero geographic exclusions — self-host it, fine-tune it, ship it in commercial products, anywhere.\n\nThat matters more than it sounds. GLM-5.2 outperforms Hy3 on SWE-bench, but carries usage constraints that matter to enterprise deployments. Hy3 with a clean Apache 2.0 license at $0.20/$0.80 per million tokens post-July-21 is a different risk profile for teams that can’t accept third-party data egress or unclear regional terms.\n\n## Self-Hosting: What You Actually Need\n\nThe weights are on Hugging Face at [tencent/Hy3](https://huggingface.co/tencent/Hy3) (BF16) and [tencent/Hy3-FP8](https://huggingface.co/tencent/Hy3-FP8). Self-hosting a 295B MoE model isn’t casual — Tencent recommends 8 H20 GPUs or equivalent high-memory cards.\n\nFor vLLM, the relevant flags are `--tool-call-parser hy_v3`\n\nand `--reasoning-parser hy_v3`\n\nwith `--enable-auto-tool-choice`\n\n. SGLang uses `--tool-call-parser hunyuan`\n\nand `--reasoning-parser hunyuan`\n\n. Both support MTP speculative decoding for faster inference.\n\nMost teams will start with the OpenRouter API and self-host only if the use case justifies the GPU spend. That’s a reasonable progression.\n\n## What to Do Before July 21\n\nIf you run agent workflows involving web search, MCP tools, or long-context retrieval: Hy3 is the right thing to test this week. Its SWE-bench score is real but not the whole picture — the tool and search benchmarks suggest a model tuned for exactly this class of workload.\n\nIf you need the best pure coding performance in the open-weight tier: GLM-5.2 still wins. Hy3 is not that model.\n\nStart at [OpenRouter’s free endpoint](https://openrouter.ai/tencent/hy3:free). The weights are at [github.com/Tencent-Hunyuan/Hy3](https://github.com/Tencent-Hunyuan/Hy3). The free window closes July 21.", "url": "https://wpnews.pro/news/tencent-hy3-295b-moe-hits-swe-bench-78-free-api-ends-july-21", "canonical_source": "https://byteiota.com/tencent-hy3-295b-moe-hits-swe-bench-78-free-api-ends-july-21/", "published_at": "2026-07-08 07:10:13+00:00", "updated_at": "2026-07-08 07:40:04.321673+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-tools", "ai-products", "ai-infrastructure"], "entities": ["Tencent", "Hy3", "OpenRouter", "Claude Sonnet 5", "GLM-5.2", "Hugging Face", "vLLM", "SGLang"], "alternates": {"html": "https://wpnews.pro/news/tencent-hy3-295b-moe-hits-swe-bench-78-free-api-ends-july-21", "markdown": "https://wpnews.pro/news/tencent-hy3-295b-moe-hits-swe-bench-78-free-api-ends-july-21.md", "text": "https://wpnews.pro/news/tencent-hy3-295b-moe-hits-swe-bench-78-free-api-ends-july-21.txt", "jsonld": "https://wpnews.pro/news/tencent-hy3-295b-moe-hits-swe-bench-78-free-api-ends-july-21.jsonld"}}