AITOPAI热报

Kimi K2.6

Moonshot AI·China·Released 2026-04-20
Open weightsGeneral LLMReasoningTool useVision
256Kcontext
$0.95 / $4per 1M tokens (in / out)
1.03Tparameters
14stories in our feed

Specs

hover a value to see its source
Context window256K
Max output256K
Parameters1.03T
Knowledge cutoff2025-01
Licenceother
Commercial useRestricted
InputText · Image · Video
OutputText
HF downloads (30d)237.2K
Familykimi-k2

API price

USD per 1M tokens
Input$0.95
Output$4
Cached input$0.16

No price change recorded since we started tracking.

News timeline

from the AITOP feed, one story per event
2026-10-22
Google DeepMind 论文：RL 训练后模型被观察与不被观察时的合规差距扩至 43%-95%rohanpaul_ai
2026-10-20
基于 Kimi K2.6 的研究显示 RL 训练会放大模型的策略性欺骗行为arXiv: Anthropic
2026-10-15
论文提出 Memory Echo 现象与 FPA 框架，修复智能体长期记忆失真arXiv: DeepSeek
2026-10-13
CoreWeave Forge 将投机解码优化流程整合到单一环境CoreWeave
2026-09-23
CliffCompaction：让长程编程智能体的上下文压缩成本降低最多 50%arXiv cs.LG
2026-08-17
Sakana Namazu上线OpenRouter，基于Kimi K2.6的日本文化推理模型OpenRouter
2026-08-11
开源模型能否媲美闭源模型？金融文本理解基准更新arXiv: DeepSeek
2026-08-01
DoorDash用月之暗面Kimi K2.6写代码 遭美众议院委员会调查IT之家
2026-07-17
Kimi-K3以1679分登顶Frontend Code Arena，超越Claude Fable 5AI Will
LM Studio Bionic 智能体发布：可调用 GLM 5.2 等模型处理编程任务IT之家
2026-07-16
Kimi K3在Text Arena排名第9，超越Kimi K2.6lmarena.ai
近4个月AI模型发布时间线一览Suhail
2026-07-13
Novita Labs 开源 DSpark 投机解码器，加速 Kimi-K2.6 与 K2.7-CodevLLM
2026-07-07
DoorDash 研究：用 Kimi K2.6 和 Claude Fable 5 搭配提升代码审查效率The Rundown AI