Compare / Qwen3.6 35B-A3B vs Kimi Linear 48B-A3B

Qwen3.6 35B-A3B vs Kimi Linear 48B-A3B

Updated November 2026

Verdict

Qwen3.6 35B-A3B scores higher on the Small LM Index: 68.3 vs 36.6. On the 8 benchmarks both labs report, Qwen3.6 35B-A3B wins 8 and Kimi Linear 48B-A3B wins 0. Qwen3.6 35B-A3B is smaller: it needs about 22.6 GB of memory at 4-bit, against 30.8 GB for Kimi Linear 48B-A3B. Only Qwen3.6 35B-A3B has a hosted API on OpenRouter ($0.031 per million tokens); the other has to be self-hosted.

Pick Qwen3.6 35B-A3B if…

you want the strongest model overall
you mostly write code
you need math and reasoning
you need precise instruction following
you build agents that call tools
you have limited memory (about 22.6 GB at 4-bit)
you want a hosted API rather than running it yourself
you need image input
you want the more widely used model (more quantized builds and community support)

Pick Kimi Linear 48B-A3B if…

you work with long documents

Specifications

	Qwen3.6 35B-A3B	Kimi Linear 48B-A3B
Small LM Index	68.3 (likely 67–70)	36.6 (likely 34–40)
Lab	Alibaba (Qwen) · China	Moonshot AI (Kimi) · China
Released	2026-04-16	2025-10-30
Parameters	35B (3B active, MoE)	48B (3B active, MoE)
Memory at 4-bit	~22.6 GB	~30.8 GB
Context window	262K tokens	1M tokens
Input	text, image, video	text
Reasoning	Hybrid (thinking on/off)	No thinking
License	Apache-2.0	MIT
API price / 1M tokens	$0.031 ($0.010 in, $0.092 out)	Self-host only
Hugging Face downloads (30 days)	4.5M	69K

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	Qwen3.6 35B-A3B	Kimi Linear 48B-A3B
MMLU-Pro Knowledge	85.2	58.5^✓
GPQA Diamond Knowledge	84.1^✓	41.2^✓
Humanity's Last Exam Reasoning	22.2^✓	2.5^✓
AIME Math	92.7	36.3^✓
LiveCodeBench Coding	80.4	37.8^✓
SWE-bench Verified Coding	73.4	≈ 40
Terminal-Bench Agentic	51.5	—
τ²-Bench Agentic	95.3^✓	0.0^✓
IFBench Instruction following	64.4^✓	28.1^✓
MMMU-Pro Vision	75.3	—
AA-LCR Long context	71.7^✓	28.0^✓
SciCode Coding	36.6^✓	—

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, Qwen3.6 35B-A3B or Kimi Linear 48B-A3B?

Qwen3.6 35B-A3B scores higher on the Small LM Index: 68.3 vs 36.6. On the 8 benchmarks both labs report, Qwen3.6 35B-A3B wins 8 and Kimi Linear 48B-A3B wins 0.

Which is cheaper to run?

Only Qwen3.6 35B-A3B has a hosted API on OpenRouter ($0.031 per million tokens); the other has to be self-hosted. Qwen3.6 35B-A3B is smaller: it needs about 22.6 GB of memory at 4-bit, against 30.8 GB for Kimi Linear 48B-A3B.

Which has the longer context window?

Kimi Linear 48B-A3B has the longer context window: 1M tokens against 262K.

Can I use Qwen3.6 35B-A3B and Kimi Linear 48B-A3B commercially?

Qwen3.6 35B-A3B is released under Apache-2.0 and Kimi Linear 48B-A3B under MIT. Apache 2.0 and MIT allow commercial use; other licenses can add conditions, so read the license before shipping a product.