Alibaba just launched Qwen3.8-Max, its most powerful flagship multimodal model featuring 2.4 trillion total parameters while activating just 95 billion via sparse MoE. It supports a 1M token context window and ranks second in Vision Arena. The model is accessible via API on Alibaba Cloud Model Studio, with weights releasing next week.
endless-frontier just launched BigBang-v1 on Hugging Face, an experimental 35B model built on an adversarial synthetic data loop. Instead of relying on human experts to write training tasks, generator and critic agents iteratively designed and solved roughly 10,000 frontier scientific challenges. Why it matters: it rivals trillion-parameter giants on research benchmarks using remarkably little data.
OpenAI just revealed that an internal version of its next major model, Astra, solved or made substantial progress on 10 open problems in math and theoretical computer science. The breakthroughs range from lattice cryptography to disproving Connes's rigidity conjecture, with all arguments verified via Lean certificates. Finding the proofs cost roughly $2,000 at Sol API rates.
DeepSeek just officially released DeepSeek-V4-Flash-0731 with built-in DSpark speculative decoding. Thanks to heavy post-training, this lightweight model leapfrogs the larger V4-Pro preview across terminal and coding benchmarks, punching well into Opus 4.8 territory while fully open-sourced under an MIT license.
The MatrAIx team just open-sourced MatrAIx-Persona-8B alongside a 1M persona dataset on Hugging Face to stress-test AI products against diverse simulated users. Spanning 1,290 psychological and behavioral dimensions across web, app, survey, and chat environments, it lets developers evaluate software at population scale before real-world deployment.
Google just plugged its Gemini Spark agent directly into Chrome to automate tedious web tasks like scheduling apartment tours or booking flights. With permission, it accesses saved passwords and active logins to navigate sites, handing control back for sensitive actions like payments. Auto-browse is rolling out first in the U.S., while Spark expands to Google AI Pro users in 160+ countries.
Frontis just open-sourced OpenRSI alongside its Frontis-MA1 model, tackling recursive self-improvement directly inside machine learning engineering. Built around four core operators—Draft, Improve, Debug, and Crossover—the 35B model on a single 12GB RTX 4090 beat GPT-5.5 + Codex on MLE-Bench Lite. It makes "AI improving AI" measurable and reproducible without mega-cluster scale.
Google DeepMind just launched Gemini Robotics ER 2, an embodied reasoning model designed to act as a high-level brain for robots. It tracks tasks via live video with 91.3% moment-finding accuracy, plans next steps while moving, and coordinates multiple robots at once. Why it matters: it eliminates the awkward "stop-and-think" pauses that keep physical AI stuck in labs.
Thinking Machines just launched Inkling-Small, an open-weights MoE model with 276B total parameters and only 12B active. Operating at roughly a quarter of the compute cost, it actually edges out the flagship Inkling across reasoning and agentic coding benchmarks like Humanity's Last Exam. Why it matters: You get multimodal reasoning and a 1M context window without the heavyweight inference tax.
OpenAI just dropped API pricing across the GPT-5.6 lineup, slashing Luna by 80% down to $0.20 per million input tokens and Terra by 20%, alongside a new 2.5x faster Fast mode for Sol. The massive price drop makes multi-step agentic workflows and high-volume data processing drastically cheaper to run at enterprise scale.
Perplexity just upgraded Spaces into Projects, giving users a persistent hierarchical file system and a self-improving memory engine called Brain. Humans and agents can now edit shared files simultaneously without starting from scratch, while pulling context from over 400 integrations like Google Drive and GitHub. Existing Spaces migrate automatically, turning Perplexity into a true shared workspace.
Tavus just launched PAL Maker, a no-code platform that lets anyone build interactive AI video companions simply by describing them to a creator agent named Charlie. Powered by models like Raven-1 and Phoenix-4, the avatars come equipped with faces, voices, vision, and long-term memory. Why it matters: It lowers the barrier of digital human creation to a casual chat, already serving roles from post-surgery check-ins to candidate screening.
YC just open-sourced QM, a multiplayer agent harness designed for company collaboration. Instead of treating AI as an isolated personal assistant, QM gives every teammate a personal sandbox and scoped memory, while letting groups collaborate with the agent in Slack channels and projects across granular permission layers.
OpenAI found that GPT-5.6 Sol's surprisingly low ARC-AGI-3 benchmark score was caused by the test harness dropping reasoning history and aggressively truncating context. By switching to its Responses API and enabling retained reasoning alongside compaction, the model tripled its score while slashing output tokens by 6x. It is a stark reminder that benchmarks evaluate harness architecture just as much as raw intelligence.
Google just rolled out major updates to Gemini API Managed Agents, making Gemini 3.6 Flash the new default model. Developers can now run custom pre- and post-tool scripts inside remote sandboxes to lint or block actions, set hard token limits to stop runaway costs, and test autonomous workflows on a free tier.
ReDesign reverses flat raster images back into editable layer hierarchies with typography, vector shapes, and depth order. An agentic vision controller orchestrates tools like Qwen-Image-Layered and SAM 2 to recursively unpack elements, and the code is now open-source on GitHub.
Perplexity just launched Personal Computer for Windows, bringing its autonomous agent out of the browser and straight into your local desktop environment. It can directly edit local Word and Excel docs, organize messy downloads, and pull real-time web and enterprise data into local workflows. It is available now for Pro, Max, and Enterprise subscribers.
Kimi just open-sourced Kimi K3, a massive 2.8-trillion parameter MoE model featuring native vision understanding and a 1-million-token context window. The team claims a new architecture delivers 2.5x the intelligence per unit of compute, and they are also releasing the full underlying stack, from custom attention kernels to scalable agent infrastructure.
NVIDIA, Microsoft, Hugging Face, and dozens of industry leaders just launched the Open Secure AI Alliance to build and share open-source cybersecurity defenses for the agentic era. NVIDIA is contributing its new NOOA agent framework on GitHub alongside open models. The alliance argues that defenders need transparent, locally runnable tools rather than relying solely on opaque, closed-source systems.
Alibaba just open-sourced qwen-audio-agent, a full-duplex voice runtime bridging conversational audio with background task agents. Users can converse and interrupt naturally while backend agents like OpenCode or Claude Code handle long-running code edits and report back when finished.
