Meituan’s LongCat-2.5 AI Model Was Trained Only on Chinese Chips
Meituan's LongCat-2.5 is a 1.6tn-parameter multimodal model trained on Chinese chips. Its own benchmarks show wins and clear losses against Claude.
Latest
Swiss Lab Giotto Releases Alto and Solo Models as Sovereign AI
Lausanne's Giotto launched Alto, an eight-GPU reasoning model it offers free, and single-GPU Solo, pitching European AI that avoids US cloud services.
DiffuSpace Releases Clover, an Open-Weight 35B Diffusion Model
DiffuSpace, whose team built Dream 7B at HKU, has open-sourced Clover, a 35B diffusion model it says beats Qwen3.6 on agent and coding tests.
Moonshot’s Own Tests Put Kimi K3.1 Pro at 65.7% on Terminal-Bench
Moonshot AI's Kimi K3.1 Pro is live with a 1M-token context. Its own tests put it near Claude Opus 5.5 on coding; open weights are due by 4 November.
Bitcoin Core Ends Waits of Up to a Year for Bugs LLMs Can Easily Find
Bitcoin Core will ship fixes for bugs a simple LLM prompt could find to every maintained version at once, skipping embargoes of up to a year.
Malicious LiteLLM Release on PyPI Stole Credentials, Researchers Say
LiteLLM 1.107.0 on PyPI carried a credential stealer and backdoor, SafeDep says. LiteLLM is treating it as SEV-0 and says to pin version 1.106.2.
DeepSeek Open-Sources DeepInfer, the Engine Behind Its Own Services
DeepSeek has released DeepInfer, the inference engine behind its V4.1 Flash service, under Apache 2.0. Its cache and load-balancing code is held back.
Inception Launches Mercury 3 Diffusion LLM, Claiming GPT-6 Luna Wins
Inception launches Mercury 3, a diffusion LLM with a 1M-token context. It claims wins over GPT-6 Luna and Haiku 5.5 at 4–8x the speed; unverified.
Cloudflare Says Clairic Pulls Typed Data From Text in Under a Second
Cloudflare previews Clairic, a one-pass extraction model on Workers AI at $0.10 per million input tokens. Speed and cost figures are its own tests.
MiniMax Rolls Out M3.1 Flash, a Faster, Cheaper Agentic Model
MiniMax’s M3.1 Flash targets agentic tasks with claimed near-3x speed and 70% lower cost. API now; weights later. Benchmarks are MiniMax’s own.
Indonesia Gets a 27B Open-Weight Model Tuned for Local Languages
GoTo’s 27B Gemma-based model covers Indonesian plus five regional languages. Open-weight under Google’s Gemma licence; AI Singapore co-developed.
Related stories
Latest from topics connected to LLM. See all with connected topics
Latest on TSN
Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.