Compare / Qwen3 30B-A3B Thinking 2507 vs Kanana-2 30B-A3B Thinking
Qwen3 30B-A3B Thinking 2507 vs Kanana-2 30B-A3B Thinking
Updated November 2026
Verdict
Qwen3 30B-A3B Thinking 2507 scores higher on the Small LM Index: 59.3 vs 50.7. On the 5 benchmarks both labs report, Qwen3 30B-A3B Thinking 2507 wins 5 and Kanana-2 30B-A3B Thinking wins 0. Both need about 20.4 GB of memory at 4-bit quantization. Neither model has a hosted API on OpenRouter, so you would run both yourself.
Pick Qwen3 30B-A3B Thinking 2507 if…
- you want the strongest model overall
- you mostly write code
- you need math and reasoning
- you need precise instruction following
- you work with long documents
- you need a permissive license (Apache-2.0)
- you want the more widely used model (more quantized builds and community support)
Pick Kanana-2 30B-A3B Thinking if…
No clear advantage in the data we have.
Specifications
| Qwen3 30B-A3B Thinking 2507 | Kanana-2 30B-A3B Thinking | |
|---|---|---|
| Small LM Index | 59.3 (likely 55–63) | 50.7 (likely 46–55) |
| Lab | Alibaba (Qwen) · China | Kakao (Kanana) · Rest of Asia |
| Released | 2025-07-30 | 2026-01-14 |
| Parameters | 30.5B (3.3B active, MoE) | 30.7B (3B active, MoE) |
| Memory at 4-bit | ~20.4 GB | ~20.2 GB |
| Context window | 262K tokens | 32K tokens |
| Input | text | text |
| Reasoning | Always thinks | Always thinks |
| License | Apache-2.0 | Kanana License |
| API price / 1M tokens | Self-host only | Self-host only |
| Hugging Face downloads (30 days) | 66.6K | 1.3K |
Benchmarks
Higher is better. Bold marks the winner where both labs report a score. ≈ marks the index model's estimate for a score that wasn't reported, ✓ an independent measurement and * a score measured by another lab.
| Benchmark | Qwen3 30B-A3B Thinking 2507 | Kanana-2 30B-A3B Thinking |
|---|---|---|
| MMLU-Pro Knowledge | 80.9 | 74.2 |
| GPQA Diamond Knowledge | 73.4 | 57.8 |
| Humanity's Last Exam Reasoning | 9.8 | ≈ 7 |
| AIME Math | 85.0 | 74.0 |
| LiveCodeBench Coding | 66.0 | 58.8 |
| SWE-bench Verified Coding | 22.0* | ≈ 32 |
| IFBench Instruction following | 51.5 | 47.8 |
| IFEval Instruction following | — | 82.2 |
Scores are mostly reported by each lab with its own settings, so small gaps are noise. See the methodology. Open in the interactive comparison to add more models.
Questions
Which is better, Qwen3 30B-A3B Thinking 2507 or Kanana-2 30B-A3B Thinking?
Qwen3 30B-A3B Thinking 2507 scores higher on the Small LM Index: 59.3 vs 50.7. On the 5 benchmarks both labs report, Qwen3 30B-A3B Thinking 2507 wins 5 and Kanana-2 30B-A3B Thinking wins 0.
Which is cheaper to run?
Neither model has a hosted API on OpenRouter, so you would run both yourself. Both need about 20.4 GB of memory at 4-bit quantization.
Which has the longer context window?
Qwen3 30B-A3B Thinking 2507 has the longer context window: 262K tokens against 32K.
Can I use Qwen3 30B-A3B Thinking 2507 and Kanana-2 30B-A3B Thinking commercially?
Qwen3 30B-A3B Thinking 2507 is released under Apache-2.0 and Kanana-2 30B-A3B Thinking under Kanana License. Apache 2.0 and MIT allow commercial use; other licenses can add conditions, so read the license before shipping a product.
Related comparisons
- Muse Glimmer 30B vs Qwen3 30B-A3B Thinking 2507
- Gemma 4 31B vs Qwen3 30B-A3B Thinking 2507
- Granite 4.2 30B vs Qwen3 30B-A3B Thinking 2507
- GLM-4.7-Flash vs Qwen3 30B-A3B Thinking 2507
- Kanana-2 30B-A3B Thinking vs Muse Glimmer 30B
- Gemma 4 31B vs Kanana-2 30B-A3B Thinking
- Granite 4.2 30B vs Kanana-2 30B-A3B Thinking
- GLM-4.7-Flash vs Kanana-2 30B-A3B Thinking