Qwen 3.6 Plus Preview matters right now for one reason: OpenRouter is exposing qwen/qwen3.6-plus-preview:free at $0 with a 1,000,000-token context window.
That makes it interesting for value hunters. It does not make it a settled winner, a verified local-model release, or a benchmark-backed production default.
Quick verdict
| Question | Answer |
|---|---|
| Worth testing? | Yes, if you want a free long-context eval path |
| Best current use | Codebase review, large-doc analysis, prompt evals, low-cost experimentation |
| Best current value proposition | 1M context at $0 while the preview lasts |
| Biggest catch | No first-party Qwen 3.6 benchmark sheet or system card was found on checked official surfaces |
| Local/offline choice? | No verified local or open-weight Qwen 3.6 release was found; treat this as a hosted preview |
| Production-safe default? | No. Use it for evaluation first, not for privacy-sensitive or repeatability-critical workloads |
What is verified as of 2026-04-02
| Item | Verified status | Source |
|---|---|---|
| Model ID | qwen/qwen3.6-plus-preview:free | OpenRouter model page |
| Release date shown by host | March 30, 2026 | OpenRouter model page |
| Context window | 1,000,000 tokens | OpenRouter model page |
| Preview price | $0/M input, $0/M output | OpenRouter model page |
| Reasoning support | OpenRouter shows reasoning support and reasoning tokens in responses | OpenRouter model page |
| Data handling warning | OpenRouter says prompts and completions may be used to improve the model | OpenRouter model page |
| Free-model rate limit | :free models are limited to 20 requests/minute | OpenRouter limits docs |
| Free-model daily limit | 50/day for accounts with under 10 credits purchased, 1000/day after purchasing at least 10 credits | OpenRouter limits docs |
What is still not verified
- No first-party Qwen 3.6 benchmark sheet or system card was found on checked official Qwen surfaces: qwen.ai, QwenLM on GitHub, and Qwen on Hugging Face.
- No verified first-party detail was found for Qwen 3.6 max output, latency, tool-calling behavior, multimodal behavior, or post-preview pricing.
- No verified local or open-weight Qwen 3.6 release was found on those official surfaces as of April 2, 2026.
Why value hunters should care
- Free long-context evals: 1M context at $0 is rare and useful for codebase review, giant prompts, and large-doc testing.
- Low-friction comparison: You can run your eval set against Qwen without committing to a paid provider first.
- Good fit for the site’s stack logic: it complements OpenCode as a provider-neutral coding harness and Gemini 3 Flash as the published long-context production value option.
The important distinction: this page is about cheap evaluation leverage, not a proven long-term default.
Benchmarks: keep the anchor conservative
Until Qwen publishes first-party 3.6 benchmark material, the safest benchmark anchor is official Qwen 3.5 family data:
| Benchmark | Qwen3.5-397B-A17B | Why it matters | Source |
|---|---|---|---|
| SWE-bench Verified | 76.4 | Coding baseline | Qwen 3.5 model card |
| GPQA | 88.4 | Reasoning baseline | Qwen 3.5 model card |
| LongBench v2 | 63.2 | Long-context baseline | Qwen 3.5 model card |
| HLE with tool use | 48.3 | Tool-augmented baseline | Qwen 3.5 model card |
Use that as family context only. It is not proof that Qwen 3.6 matches or exceeds those numbers.
Local model reality
If your question is “is this the local-model champion right now?”, the current answer is no, not on verified evidence in this page.
What is verified today is a hosted preview on OpenRouter. What is not yet verified is a first-party Qwen 3.6 local or open-weight release with published benchmarks and deployment docs. If you want a documented Qwen family anchor for local or self-hosted evaluation, Qwen 3.5 is still the safer reference point.
Access and limits
| Item | Current status |
|---|---|
| Access path | OpenRouter |
| Model slug | qwen/qwen3.6-plus-preview:free |
| Endpoint style | Standard chat/completions API |
| Cost | Free while preview listing remains active |
| Throughput ceiling | 20 requests/minute on :free |
| Daily ceiling | 50/day or 1000/day depending on purchase history |
| Reliability profile | Shared-capacity preview, not a stable official vendor commitment |
OpenRouter’s own docs frame free models as experimentation and prototyping paths. That is the right way to think about Qwen 3.6 Plus Preview today.
When to use it
- When you want a free 1M-context test pass before spending money
- When you are comparing long-context behavior against Gemini 3 Flash
- When you want a Qwen-family slot in your eval stack beside Kimi k2.5
- When you are doing codebase review, prompt tuning, or document-heavy experiments
When to skip it
- When you need local or offline deployment
- When private code or sensitive data cannot touch a host that warns about model-improvement usage
- When you need benchmark-backed selection criteria for a serious rollout
- When you need stable pricing, an SLA, or clear first-party commercial commitments
OpenCode Is A Separate Path
OpenCode can run Qwen through supported providers, but its current Zen pricing table does not label Qwen rows free. Do not treat this OpenRouter preview entitlement as an OpenCode Zen entitlement.
How it compares to OpenRouter:
- OpenCode: coding harness; provider and billing path determine cost and terms
- OpenRouter: current
:freeAPI listing with free-model rate limits
Choose OpenCode when: You want a coding-agent workflow and have selected a provider path. Choose OpenRouter when: You specifically want the currently listed free preview API.
Related links
- /models/
- /value/free-stack/
- /value/smart-spend/
- /tools/opencode/ — Provider-neutral harness and current Zen billing paths
- /models/kimi-k2.5/
- /models/gemini-3-flash/