Kimi K2.6
moonshot-kimi-k2-6
fields
| nameREQ | Kimi K2.6 | www.buildfastwithai.com | 2026-05-03 |
| vendorREQ | Moonshot AI | www.buildfastwithai.com | 2026-05-03 |
| released_at | 2026-04-20 | www.buildfastwithai.com | 2026-05-03 |
| modality | text, code | www.buildfastwithai.com | 2026-05-03 |
| access | open-weights2 revisions | awesomeagents.ai | 2026-05-05 |
| context_window | 262144 | awesomeagents.ai | 2026-05-05 |
| parameters | 1T total / 32B active | awesomeagents.ai | 2026-05-05 |
| benchmark_score | 80.22 revisions | awesomeagents.ai | 2026-05-05 |
| benchmark_name | SWE-Bench Verified2 revisions | awesomeagents.ai | 2026-05-05 |
| announcement_url | https://huggingface.co/moonshotai/Kimi-K2.62 revisions | awesomeagents.ai | 2026-05-05 |
history · 10 fields · 14 revisions
name1 revision
Kimi K2.6
current
Kimi K2.6 beat GPT-5.4 on SWE-Bench Pro at $0.60 per million output tokens and achieved Tier A (87/100) on real-world coding benchmarks — the only Chinese model to reach that tier. Supports 300-agent parallel swarm orchestration.
vendor1 revision
Moonshot AI
current
Kimi K2.6 beat GPT-5.4 on SWE-Bench Pro at $0.60 per million output tokens and achieved Tier A (87/100) on real-world coding benchmarks — the only Chinese model to reach that tier. Supports 300-agent parallel swarm orchestration.
released_at1 revision
2026-04-20
current
Kimi K2.6 beat GPT-5.4 on SWE-Bench Pro at $0.60 per million output tokens and achieved Tier A (87/100) on real-world coding benchmarks — the only Chinese model to reach that tier. Supports 300-agent parallel swarm orchestration.
modality1 revision
text, code
current
Kimi K2.6 beat GPT-5.4 on SWE-Bench Pro at $0.60 per million output tokens and achieved Tier A (87/100) on real-world coding benchmarks — the only Chinese model to reach that tier. Supports 300-agent parallel swarm orchestration.
access2 revisions
open-weights
current
Moonshot AI released Kimi K2.6 on April 20 with open weights on HuggingFace under a Modified MIT license (moonshotai/Kimi-K2.6). 1T total / 32B active MoE. SWE-Bench Verified 80.2%, SWE-Bench Pro 58.6%. Context window 256K.
closed
superseded
Kimi K2.6 beat GPT-5.4 on SWE-Bench Pro at $0.60 per million output tokens and achieved Tier A (87/100) on real-world coding benchmarks — the only Chinese model to reach that tier. Supports 300-agent parallel swarm orchestration.
context_window1 revision
262144
current
Moonshot AI released Kimi K2.6 on April 20 with open weights on HuggingFace under a Modified MIT license (moonshotai/Kimi-K2.6). 1T total / 32B active MoE. SWE-Bench Verified 80.2%, SWE-Bench Pro 58.6%. Context window 256K.
parameters1 revision
1T total / 32B active
current
Moonshot AI released Kimi K2.6 on April 20 with open weights on HuggingFace under a Modified MIT license (moonshotai/Kimi-K2.6). 1T total / 32B active MoE. SWE-Bench Verified 80.2%, SWE-Bench Pro 58.6%. Context window 256K.
benchmark_score2 revisions
80.2
current
Moonshot AI released Kimi K2.6 on April 20 with open weights on HuggingFace under a Modified MIT license (moonshotai/Kimi-K2.6). 1T total / 32B active MoE. SWE-Bench Verified 80.2%, SWE-Bench Pro 58.6%. Context window 256K.
87
superseded
Kimi K2.6 beat GPT-5.4 on SWE-Bench Pro at $0.60 per million output tokens and achieved Tier A (87/100) on real-world coding benchmarks — the only Chinese model to reach that tier. Supports 300-agent parallel swarm orchestration.
benchmark_name2 revisions
SWE-Bench Verified
current
Moonshot AI released Kimi K2.6 on April 20 with open weights on HuggingFace under a Modified MIT license (moonshotai/Kimi-K2.6). 1T total / 32B active MoE. SWE-Bench Verified 80.2%, SWE-Bench Pro 58.6%. Context window 256K.
Real-World Coding (Tier A score /100)
superseded
Kimi K2.6 beat GPT-5.4 on SWE-Bench Pro at $0.60 per million output tokens and achieved Tier A (87/100) on real-world coding benchmarks — the only Chinese model to reach that tier. Supports 300-agent parallel swarm orchestration.
announcement_url2 revisions
https://huggingface.co/moonshotai/Kimi-K2.6
current
Moonshot AI released Kimi K2.6 on April 20 with open weights on HuggingFace under a Modified MIT license (moonshotai/Kimi-K2.6). 1T total / 32B active MoE. SWE-Bench Verified 80.2%, SWE-Bench Pro 58.6%. Context window 256K.
https://kimi.ai
superseded
Kimi K2.6 beat GPT-5.4 on SWE-Bench Pro at $0.60 per million output tokens and achieved Tier A (87/100) on real-world coding benchmarks — the only Chinese model to reach that tier. Supports 300-agent parallel swarm orchestration.