Qwen3.5 397B A17B vs Hy4 preview
Benchmark Performance
Available Benchmarks
Technical Differences
Side-by-Side Facts
| Field | Qwen3.5-397B-A17B | Hy4 preview |
|---|---|---|
| Developer | Qwen | Tencent |
| Family | Qwen3 5 397b A17b | Hy4 |
| Model | Qwen3.5-397B-A17B | Hy4 preview |
| Version | Qwen3.5-397B-A17B | Hy4 preview |
| Lifecycle | active | preview |
| Released | 2026-02-15 | 2026-08-28 |
| Knowledge cutoff | Unknown | Unknown |
| Input modalities | Text, Image | Text |
| Output modalities | Text | Text |
| Context window | 262K | 1,000K |
| Total parameters | 403.4B | 770B |
| Active parameters | 17B | 49B |
| License | apache-2.0 | apache-2.0 |
| Open weights | Yes | Yes |
| API available | Yes | Yes |
| Self-hostable | Yes | Yes |
| Provider access | Deepinfra (Standard), Hugging Face (Standard), Openrouter (Standard), Together Ai (Standard) | Openrouter (Standard) |
| Capabilities | chat, generation, reasoning, tools | chat, generation, reasoning, tools |
17 comparable fields · 11 material differences · Pair passes the primary-source comparison gate
Qwen3.5 397B A17B Capabilities
Hy4 preview Capabilities
Internal Comparison Graph
Related Comparisons
| A | Pair | B | Context |
|---|---|---|---|
| vs |  GPT-6 AstraOpenAI | cross-developer peersimage, text | |
| vs | Claude Fable 5.1Anthropic | cross-developer peersimage, text | |
| vs |  Gemini 3.1 ProGoogle DeepMind | cross-developer peerstext | |
| vs |  DeepSeek-V4-ProDeepSeek | cross-developer peerstext | |
| vs | Grok 4.6xAI | cross-developer peersimage, text | |
| vs | Qwen3.8-MaxQwen | family variantsimage, text | |
| vs |  Kimi-K3Moonshot AI | cross-developer peersimage, text | |
 MiniMax-M3MiniMax | vs | cross-developer peersimage, text | |
| vs |  GLM-5.3Z.ai | cross-developer peerstext | |
| vs | Seed 2.1 ProByteDance Seed | cross-developer peersimage, text | |
| vs |  Mistral Large 3Mistral AI | cross-developer peersimage, text | |
| vs | cross-developer peersimage, text |
Primary Evidence
Sources and Freshness
Questions
Qwen3.5 397B A17B vs Hy4 preview FAQs
Is Qwen3.5 397B A17B or Hy4 preview better for coding?+
This comparison does not currently contain a protocol-matched coding benchmark for both Qwen3.5 397B A17B and Hy4 preview, so Model Markets cannot name a coding leader from pricing, context size, or capability labels alone.
Which is cheaper, Qwen3.5 397B A17B or Hy4 preview?+
Qwen3.5 397B A17B is $0.45 and Hy4 preview is $0.834 per million tokens, so Qwen3.5 397B A17B is cheaper on this metric. Qwen3.5 397B A17B is $3.00 and Hy4 preview is $2.501 per million tokens, so Hy4 preview is cheaper on this metric.
Which has a larger context window, Qwen3.5 397B A17B or Hy4 preview?+
Hy4 preview has the larger sourced context window. Qwen3.5 397B A17B supports 262K and Hy4 preview supports 1,000K.
Which performs better in benchmarks, Qwen3.5 397B A17B or Hy4 preview?+
There is no overall benchmark winner: At least two independently verified, protocol-matched benchmarks are required for an overall winner.
Can Qwen3.5 397B A17B or Hy4 preview be self-hosted?+
Both models have the same recorded self-hosting status: supported. Qwen3.5 397B A17B is open weight; Hy4 preview is open weight.
Can Qwen3.5 397B A17B and Hy4 preview understand images?+
Qwen3.5 397B A17B is documented with image input; Hy4 preview is not documented with image input. This reflects supported input modalities, not vision quality.
Which can generate longer answers, Qwen3.5 397B A17B or Hy4 preview?+
Neither has a larger sourced maximum output. Qwen3.5 397B A17B is — and Hy4 preview is —.
Do Qwen3.5 397B A17B and Hy4 preview support reasoning and tool use?+
Qwen3.5 397B A17B: reasoning, tool calling, and image input. Hy4 preview: reasoning and tool calling. Feature support does not establish relative quality.
Which is available from more inference providers, Qwen3.5 397B A17B or Hy4 preview?+
Qwen3.5 397B A17B has 4 sourced provider routes; Hy4 preview has 1, so Qwen3.5 397B A17B has broader tracked availability.
Which offers better value, Qwen3.5 397B A17B or Hy4 preview?+
There is no universal value winner. Compare the input and output prices above with the matched benchmark result for your workload: cheaper tokens can be offset by different quality, token usage, latency, or provider availability.