A ~501B-parameter MoE with 23B active parameters and 1M context, focused on coding and long-horizon agentic work, released under Apache 2.0 and trained on Together AI GPU clusters. Day-0 SGLang production support is live on Together, with the LMSYS team publicly congratulating the launch.
RMB 1.625 billion (~$243M) in financing to commercialize the Mach M100 (1,280 TOPS) beyond EVs. The round includes RMB 1B from Li Auto's controlling investment entity plus RMB 625M from external investors including BlueRun Ventures and MiraclePlus.
Introduces multi-session agentic RL for trajectories spanning context compaction and subagent forking, with critic-free hierarchical advantage estimation weighting sessions by marginal contribution. Beats trajectory-level advantage broadcasting on BrowseComp with Qwen3.5-4B while using fewer sessions.
Unifies GUI agent evaluation, data collection and online RL across mobile, computer and web environments at up to 20,000 parallel instances. MAI-Gym-9B hits 78.4% on MobileWorld and 76.2% on OSWorld-Verified, with mocked-app RL transferring to real phones (28.8% to 48.1%).
First release of an open generalist computer-use agent stack combining visual and programmatic interfaces, benchmarked against GPT-6 Astra's closed system. On real macOS tasks, open models like Qwen3.8-Max use 19.4x more reasoning tokens than GPT-6 Astra.
A single diffusion LLM backbone unifies image, video and 3D generation with text via multi-codebook discrete visual tokens, topping unified models in understanding, generation and editing. A stability-aware sampling strategy accelerates image generation 4.45x, and the stack runs on both Ascend and NVIDIA.
Treats intermediate clean estimates as drafts and freezes tokens early when they match their full-step output, skipping unnecessary denoising. Matches full-step quality at ~1.7 steps/token on LongLive (over 1.5x speedup, 41.7 fps) and 1.35x on Lingbot-World 2.0.
A web tool for testing real-world inference speed of local LLMs on specific GPUs before purchasing hardware.
Generates and hosts encyclopedic entries for fictional worlds, drawing community attention on launch day.
An open-source diff viewer designed for reviewing AI-generated code changes, rethinking the classic Meld workflow for machine-written diffs.
Tree-shakes MCP server schemas and acts as a firewall, reducing context bloat and risk when connecting Claude or Codex to MCP tools.
A writeup and open-source project showing how AI agents can execute instructions never explicitly approved by the user via MCP.
The first D-Wave Advantage2 quantum system with over 4,400 qubits installed in Italy was unveiled on October 16 at Villa Saporiti during the Digital Innovation Forum 2026, opening new research and industrial opportunities.
Legislative decree n. 179/2026 completes the national AI Act framework, with fines up to €35 million or 7% of turnover for prohibited practices plus transparency and incident-reporting obligations. Only nine member states have completed their authority designations.
Politecnico di Milano's Cloud Ecosystem & Sovereignty Observatory reports 17% year-over-year growth, with AI and cybersecurity as the main drivers, forcing businesses to choose between hyperscalers and architectures with greater data control.
45% limit themselves to informal AI use driven by individual employees and 21% don't use AI at all, highlighting a persistent maturity gap between Italy's large enterprises and its small and medium businesses.
Portland, OR semiconductor systems startup building modular compute-and-memory hardware to cut data-movement costs in AI and hyperscale infrastructure.
New York-based financial operating system for cloud, AI and data spending, announced October 15.
The physical-AI startup emerged from stealth; the figure is cumulative financing rather than a new round.
Free means $0 per token today, not unlimited or unconditional: both platforms rate-limit free use and require a free API key, some OpenRouter free endpoints may train on your prompts unless you opt out in account settings, and the roster changes daily. Verified against each provider's own pricing data at generation time — but pricing changes without notice, so confirm it's still free on the provider's own page before you rely on it.
Source:
