AI Daily · Oct 24, 2026: Moonshot releases Kimi K3.1 Pro open-weight model with 1M-token context
Moonshot open-sources Kimi K3.1 Pro, a 2.8T-parameter model with 1M-token context, priced at $2/$10 per million tokens on OpenRouter; MiniMax M3.1 Flash also ships open-weight at $0.15 input.
Pick a date
-
Moonshot releases Kimi K3.1 Pro open-weight model with 1M-token context
Kimi K3.1 Pro is a 3T-class open-weight multimodal MoE with 2.8T total and 104B active parameters, 1M-token context, post-trained from K3 for long-horizon coding and agentic work, targeting full-project coding in Unreal, Unity and Godot plus Blender/CAD modeling. The API and a 2-3x faster Highspeed tier are live on OpenRouter, Requesty, Command Code, Qoder and OpenCode Go; OpenRouter lists $2/$10
-
MiniMax M3.1 Flash open weights released, live on multiple platforms day one
MiniMax-M3.1-Flash ships open-weight with 428B total and 23B active parameters, text/image/video input and 1M context, scoring 73.7% on DeepSWE v1.1 and 61.7 on OSWorld V2. It is live on day one via SiliconFlow, GMI Cloud (through OpenRouter), Fireworks serverless preview and Tenstorrent Cloud. Pricing is $0.15 per 1M input tokens, $0.015 cached input and $0.60 output.
-
OpenAI red-teaming finds unauthenticated RCE flaw in Brotli library
During a red-teaming exercise, OpenAI found a vulnerability allowing unauthenticated remote code execution in Brotli, a compression library used in Node.js, Bun and Envoy, letting attackers run code via malicious compressed data.
-
LiteLLM completes investigation into v1.107.0 release attack chain
LiteLLM published the findings of its investigation into the attack chain behind the v1.107.0 release, detailing how the release was compromised.
-
Cline launches $9.99/month ClinePass for discounted open-weights model access
Cline introduced ClinePass, a $9.99/month subscription giving roughly 5x discounted access to open-weights models.
-
DeepSeek V4.1-Flash-1017 free in Qoder extended to Nov 1
Qoder extended the zero-credit free window for DeepSeek-V4.1-Flash-1017 from Oct 24 to Nov 1, 00:00 UTC+8, for both free and paid users.
-
Codex mid-conversation reasoning changes no longer invalidate cache
OpenAI's Tibo reports Codex calls connect faster, changing reasoning mid-conversation no longer invalidates the cache, and the Codex CLI was updated.
-
Tencent Hunyuan releases Hy-Nomas open-source interactive video world model
Hy-Nomas is an open-source free-form interactive video world model supporting text prompts and camera trajectories, long-horizon generation with world memory, LoRA adaptation for domains like robotics, and inference on consumer GPUs.
-
NVIDIA Nemotron 3.5 Bolt models land on Ollama
NVIDIA's Nemotron 3.5 Bolt is now on Ollama for fast local chat, image understanding and tool use with a 128K-token context; the 4GB Bolt E4B variant is also available for local assistants, interactive characters and tool-using robots.
-
OpenAI removes live internet access for frontier models during training
OpenAI says it has hardened its research training environment over recent months, cutting live internet access for all frontier models and replacing it with an offline cached web for browsing, search and app use.