AI Daily · Nov 5, 2026 · Thursday
BAD INFLUENCE
Model releases & updates
Kimi.ai@Kimi_MoonshotOfficial4 more sourcesMoonshot AI releases the model weights of Kimi K3.1 Pro
Kimi, Moonshot AI's account, says it is releasing the model weights of Kimi K3.1 Pro, linking to the Hugging Face repository. The post says the company will keep advancing open intelligence, and it gives no benchmark or licensing details.
Cognition Blog (Devin, Windsurf)Official2 more sourcesCognition introduces SWE-2 Flash, a coding model with weights withheld
Cognition introduces SWE-2 Flash, a coding model it calls fully American, available today in Devin Desktop and CLI at one-fifth the price and 3x the speed of SWE-2. Cognition says Terminal-Bench 4.0 scores rose from 0.9% to 20.0% over the base model Inkling-Small, and the company is not publicly releasing the weights because of cyber misuse risk, instead planning a program to share them with trusted industry partners and research labs.
PandailyNews4 more sourcesTencent releases Hy4 with vision and computer use as an open-weight model
Tencent released Hy4, its latest large language model, which adds vision and computer use for the first time and will have its weights published shortly. The model keeps the same API pricing as Hy4 preview, at $0.834 per million input tokens and $2.501 per million output tokens. Tencent says a considerable gap remains between Hy4 and leading closed-source models.
Product launches & updates
vLLM BlogOfficialvLLM adds day-0 support for Kimi K3.1 Pro on NVIDIA, AMD, TPU and T-Head
vLLM says it supports Moonshot AI's Kimi K3.1 Pro on day 0 across NVIDIA GPUs, AMD GPUs, Google TPUs and Alibaba T-Head PPUs. The post says the model adds Sparse Attention (DSA) to all 24 full-attention layers, and that vLLM packs its MLA latent, indexer keys and KDA state into one KV cache. vLLM also reports up to 1.69× faster time to first token than head-sharded prefill at 340K tokens on 8× GB300.
Databricks BlogOfficialApache Spark 4.3 adds stateful real-time processing and a quarterly release cadence
Apache Spark 4.3 extends Real-Time Mode to stateful workloads, with P99 end-to-end latency below 150 ms in three tested benchmarks. The release also adds SCD Type 2 Auto CDC, Spark Connect observability features, and a nanosecond timestamp preview, and Spark moves to quarterly feature releases.
meng shao@shao__mengXTencent's Hy4 is free in WorkBuddy from November 4 to 15
Tencent's Hy4 model is free to use in WorkBuddy from November 4 to 15, according to the author's post and screenshot. The quoted Tencent announcement describes Hy4 as the first Hunyuan model with vision and Computer Use, with weights promised but not yet released.
Rohan Paul@rohanpaul_aiXAxiom Math releases a Lean formalization of the Stacks Project
Axiom Math says it has released a 16-million-line Lean formalization of the Stacks Project, a reference textbook on algebraic geometry. Lean checks each completed proof, and each tagged result links to its original text and dependencies. The library is described as free and open source, so mathematicians and AI models can reuse verified definitions and theorems.
Industry news
OpenAI Newsroom@OpenAINewsroomOfficial6 more sourcesOpenAI to acquire Pierre to expand Codex for team software development
OpenAI announced it will acquire Pierre, the developer tools company. OpenAI says the deal will accelerate its plan to make Codex a place where teams build software together with AI, from first idea to shipped version. According to OpenAI, Pierre's open-source developer tools already power parts of Codex, and the two companies will build on that foundation.
The Verge · AINewsMidterm voters back data center restrictions and moratorium candidates
Voters in the 2026 midterms approved restrictions on data center development and elected candidates who favor moratoriums, according to The Verge. Dozens of local ballot measures restricted development, with 33 of 45 measures restricting data centers as of Wednesday morning. NBC News found 76 percent of voters it polled opposed data centers in their communities.
Papers & research
Anthropic@AnthropicAIOfficial1 more sourceAnthropic maps the first 3D structure of a bacterial vault particle with Claude
Anthropic says its biology research team used Claude to determine the first 3D structure of a vault particle from bacterial life. Vault particles, first discovered in 1986, have a function that remains unknown, and similar proteins from bacteria and their viruses can also assemble into vaults. Anthropic says Claude cut work that would have taken weeks to months to days, and that it made some simple mistakes which its machine learning researchers helped address.
Tips & takes
meng shao@shao__mengXLovable's founder recounts rebuilding its coding agent architecture several times
Lovable founder Anton Osika says the product goal has stayed the same since 2023 while the implementation was rebuilt several times. Lovable narrowed the problem to new projects, cut agent autonomy, and replaced fine-tuning with knowledge files. It later moved to branching and merging whole running machines, and says it has built over 70 million projects and merged over 100,000 PRs.
