2026 Kimi K2.6 Long-horizon agentic coding， image， MuonClip optimizer for zero training instability across 15.5T tokens. Set the open-source agentic baseline: SWE-bench 65.8%， VideoMMU 86.6%. The breakthrough multimodal release that established Kimi's vision capabilities. 100 sub-agents · 4.5× faster Learn more → // REASONING · NOV 2025 Kimi K2 Thinking Post-trained reasoning variant with interleaved chain-of-thought and native tool use. 200–300 sequential tool calls without losing task context. Native INT4 quantization via QAT for 2× speed vs FP16 - no accuracy loss. Tencent CodeBuddy integrates K2 Thinking as its core engine. Pioneered the think→act→observe→think loop at production scale. 300 tool calls · 2× speed INT4 Learn more → // FLAGSHIP OPEN-SOURCE · JULY 2025 Kimi K2 The original trillion-parameter open-source frontier. 1T MoE， 4.5× execution speedup. SWE-bench 76.8%， 1， 32B active，。

2026 Kimi K2.5 Native multimodal intelligence trained on 15T mixed visual + text tokens. MoonViT 400M encoder for images and video. 256K context. Agent Swarm v1: 100 parallel sub-agents，000 coordinated steps. Claw Groups for cross-model collaboration. Document-to-Skill conversion. 262K context window. SWE-bench Verified 80.2% - SOTA among open-source models. BrowseComp Swarm 86.3%. Supports instant mode and thinking mode. The backbone of the Kimi platform for most users. 80.2% SWE-bench Verified 300 agents · 4。

τ²-bench 80%. Modified MIT License. Still a strong cost-efficient production model for text-only workflows. Open source · Modified MIT Learn more → ， ⚡ NEW · JUNE 15， beating Claude Opus 4.8 at 76.4. Multimodal via MoonViT 400M encoder - text， preserve-thinking across turns. MCP tool-use SOTA: 81.1 on MCP Mark Verified， 2026 · HIGHSPEED Kimi K2.7 Code HighSpeed The same Kimi K2.7 Code model - Moonshot's most capable coding model - served at extreme throughput. Up to 260 tokens per second on short-context tasks， and Business users. ⚡ 260 tok/s peak · 6× faster $1.90 / $8.00 per 1M Full guide → // CODING SPECIALIST · JUNE 12， ~180 tok/s on median coding inputs. Designed for agentic workflows where speed determines task completion time. Rolling out to Kimi Code Beta， general availability. 300-agent swarm with 4， and video input. Open weights under Modified MIT license. +21.8% Kimi Code Bench v2 −30% reasoning tokens Learn more → // CURRENT GA · APRIL 20， MMLU-Pro 73.3%， API developers， 128K context， AIME 96.1%，000 steps Learn more → // VISUAL AGENTIC · JAN 27， 2026 Kimi K2.7 Code Moonshot's most capable coding model to date. Reduces reasoning token usage by ~30% vs K2.6 while improving scores on every benchmark. Mandatory thinking mode。

500 tool calls。