Qwen3.6-Plus: benchmarks and LLM speed

Qwen3.6-Plus: Alibaba's Flagship LLM with 1M Token Context

Qwen3.6-Plus debuted on OpenRouter overnight from March 30–31, 2026. The model launched in free preview mode without a press release—just a tweet from a Qwen researcher featuring a benchmark table. In the first two days, it processed 400 million completion tokens across 400,000 requests. The default context window is 1 million tokens.

This makes the model suitable for tasks involving long data sequences, where standard limits of 128K–512K tokens often restrict analysis of large codebases or documents.

Benchmarks: Leadership in Terminal and Document-Oriented Tests

On Terminal-Bench 2.0, Qwen3.6-Plus scores 61.6%, surpassing Claude 4.5 Opus (59.3%). On OmniDocBench v1.5, it achieves 91.2% versus 87.7% for Claude. However, on SWE-bench Verified, it trails with 78.8% against 80.9%.

| Benchmark | Qwen3.6-Plus | Claude |

|-------------------|--------------|--------------------|

| Terminal-Bench 2.0 | 61.6% | 59.3% |

| OmniDocBench v1.5 | 91.2% | 87.7% |

| SWE-bench Verified| 78.8% | 80.9% |

The data highlights strengths in handling terminal commands and documents, but the shortfall in verified SWE tasks calls for further refinement.

Performance and User Tests

The OpenRouter community reports generation speeds 2–3 times higher than Claude Opus 4.6 in tokens/sec. This is crucial for real-time development: code autocompletion, chatbots, and agent systems benefit from low latency.

Example scenario: In IDE plugins with a Claude-like interface, Qwen3.6-Plus accelerates iterations on large projects through its combination of long context and high throughput.

Integration and Availability

The model integrates with:

• Alibaba Cloud Model Studio;
• Qwen Chat;
• Tools like Claude Code, Cline, and OpenClaw.

Free access remains available at the time of publication; pricing has not been announced. Compatibility with the Claude ecosystem simplifies migration for developers accustomed to Anthropic's API structure.

For mid/senior specialists, this enables quick A/B testing in CI/CD pipelines or agent frameworks without rewriting prompts.

Key Points

• 1M token context is ideal for analyzing monolithic repositories and lengthy specifications;
• Outperforms Claude on Terminal-Bench and OmniDocBench, but lags on SWE-bench;
• Speed 2–3x higher than Claude Opus 4.6 in tokens/s;
• Free preview on OpenRouter with 400M tokens processed in 2 days;
• Integration with Claude-compatible tools with no API changes required.

— Editorial Team