KIMI K2.6
Open SourceMoonshot AI•Released on 2026-04-13
Moonshot AI's latest open-weight flagship released April 13, 2026. A 1T-parameter MoE with 32B active per token and 256K context. Can dynamically scale to 300 sub-agents executing 4,000 coordinated steps, supporting 12-hour coding sessions. Outperforms GPT-5.4 and Claude Opus 4.6 on several coding benchmarks while being 5-6x cheaper than Sonnet 4.6.
82
Overall Score
Voice of the community
sample 180“Kimi K2.6 outperforms GPT-5.4 and Claude Opus 4.6 across several AI benchmarks. 80.2% on SWE-Bench Verified.”
“Kimi K2.6 did not release a coding model — it opened the control room. Agent horizontal scaling to 300 sub-agents across 4000 coordinated steps.”
Core Specs
256K
Context Window
0K
Max Output
ReasoningOpen Sourcetext
Scenario Scores
Pros & Cons
Sentiment75% +20% ·5% −
Pros
- +Open-weight from day 1
- +Leads SWE-Bench Verified at 80.2%
- +5-6x cheaper than Claude Sonnet 4.6
- +300 parallel sub-agents (3x K2.5)
- +12-hour autonomous coding sessions
Cons
- −Text only, no multimodal yet
- −1T params require serious GPU infrastructure to self-host
- −Community data still limited (released just days ago)
Pricing
Input (per 1M tokens)$0.60
Output (per 1M tokens)$2.50
Free trial available
Updated on 2026-04-21
Get Started
1Visit the provider's website
2Create an account
3Start using the model
Benchmarks
sweBenchVerified80.2%
sweBenchPro58.6%
terminalBench266.7%
deepSearchQA_F192.5%
Reliability
Incidents (30d)0
View Status Page →