AI — Page 26
Models, agents, infra, applied AI.
- MiniMax unveils M3, an open-weights model touting coding-agentic gains and 1M context
In a thread on X, MiniMax cites 59.0% on SWE-Bench Pro and a new sparse attention scheme that scales context to 1M, with an API promo rolling out this week.
- Agent Cookie Lets AI Agents Stay Logged In Across Your Apps
Open-source, peer-to-peer cookie and token sync pairs two Macs so agents like OpenClaw and Hermes can stay authenticated for carts, orders, and API calls.
- Insane Mario-like demo game shows the power of Opus 4.8
In a 4-post thread, Nakajima shared a demo labeled Opus 4.8 and said he will run an AI security session for non-engineers, calling out Claude Code users.
- Artisan CEO Jaspar Carmichael-Jack says KC Green dispute is settled, pulls Ava ads using 'This is fine'
Artisan took down New York and San Francisco ads that riffed on KC Green’s meme; Green says the settlement came together quickly, per TechCrunch, with terms undisclosed.
- NousResearch says it's a big week for Hermes Agent; X takes notice
A one-line tease on X pulled in 850 likes and 61 replies with no technical details.
- Apify hosts biggest SF AI hackathon for 180 builders; 80+ spots already claimed
The event centers on AI automation and web scraping projects built on Apify's platform; capacity is 180 with 80+ registrations already claimed, per Apify's post on X.
- Bankr hackathon coming after Base MCP launch
A post on X says the upcoming Bankr event will focus on agent-powered swaps and trading on Base, positioning builders to tap Coinbase’s L2 for distribution.
- Hermes Agent claims No. 1 on OpenRouter as agents crowd weekly AI usage board
OpenRouter's opt-in usage data puts Hermes Agent at No. 1; OpenClaw, Kilo Code, Descript, pi, Janitor AI, GitLawb, ISEKAI ZERO, and Cline round out the week's most-used tools.
- xAI’s Grok-Imagine-Video-1.5-Preview hits #1 on Arena.ai’s image-to-video leaderboard
Arena.ai reports a +52 point jump over the prior Grok-Imagine-Video (720p), edging past Seedance-2.0 and HappyHorse on its image-to-video leaderboard.
- Ruflo plugs multi-agent orchestration into Claude Code, claims 100+ specialized agents and shared memory
Posted on X, Ruflo is presented as a GitHub repo that runs multi-agent workflows inside Anthropic's Claude Code, with task results feeding a shared, self-learning SOP layer.
- Nature Biotechnology paper introduces MOLEA, a single-pass AI for multi-objective drug design
MOLEA reports simultaneous optimization of potency, selectivity, and safety in one go, challenging the usual one-property-at-a-time workflows in AI-assisted drug design.
- Addy Osmani packages Claude Code agent skills into slash commands that mirror the dev cycle
A post says the kit wraps senior-engineer patterns into slash commands like /spec, /plan, /build, /test, /review, /ship; it claims seven commands but lists six.
- Mystery company allegedly spent $500 million on Claude in one month after leaving license usage uncapped
Tom's Hardware, citing Axios, says an unnamed enterprise forgot to set usage limits on employee Claude licenses, amplifying worries that corporate AI spend is outpacing returns.
- Caelan Garrett to present ScheduleStream, a GPU-driven multi-arm planner, at ICRA 2026 in Vienna
The researcher says ScheduleStream tackles multi-arm task-and-motion planning on GPUs; the announcement surfaced via an X post amplified by Bowen Li.
- Xiaoxuan Ma shares REST3D, aiming for physically stable, visually consistent 3D from a single photo
In a post on X amplified by Bowen Li, the REST3D project teases single-image 3D scene reconstruction; no paper or code link was provided in the announcement.
- Liquid AI ships LFM2.5-8B-A1B, an 8B on-device MoE trained on 38T tokens
The on-device 8B MoE adds a 128K context, 128K vocab, and scaled pretraining to improve tool-calling on laptops, with base and post-trained weights on Hugging Face.
- NVIDIA releases NVFP4-quantized Qwen3.6-35B checkpoint on Hugging Face
4-bit NVFP4 build of Alibaba’s Qwen3.6-35B-A3B lands with Apache-2.0 license, long-context support, and vLLM instructions; NVIDIA notes it did not develop the base model.
- Gradio's "Build Small" hackathon opens on Hugging Face with $40k+ in prizes; registration closes June 3
The Hugging Face org for "Build Small" is live, with anchor sponsors OpenBMB, OpenAI, and NVIDIA; small-model constraints, two tracks, bonus-quest badges, and a two-weekend build window.
- Ollama adds OpenJarvis, a local-first personal AI from Stanford labs
Built with Stanford's Hazy Research and Scaling Intelligence labs under their Intelligence Per Watt program, OpenJarvis now runs locally via Ollama.
- Kog says it hit 3,000 tokens/s per request on standard GPUs
Kog opened a live coding playground and argues single-request decode speed, not FLOPS, is the bottleneck that matters for autonomous agents.
- X user @sang_yun_lee tees up 'sleep' for language models in new work
In a short X post, the user @sang_yun_lee hints at a periodic, recurrent cycle for LMs, framing the idea with: "Almost all animals sleep. Why don't LMs?"
- XCENA raises $135M to make memory, not compute, the center of AI performance
CEO Jin Kim and a Samsung/SK hynix veteran team rebrand MetisX as XCENA and double down on CXL computational memory with MX1, at a $570M valuation.
- AISlop ships CLI and GitHub Action to catch AI-generated code smells
AISlop is live on GitHub and the Marketplace as a CI quality gate with ai-slop/* rules, signaling a push to flag AI-written code in CI workflows.
- AI drug discovery enters physical validation: Insilico, Recursion, FutureHouse and Google DeepMind credited with new milestones
Peer-reviewed data for AI-generated peptides, enzymes and CRISPR variants, plus AI-designed compounds entering human trials, push in silico models into preclinical reality.