Open-Source AI
Open models — Llama, Mistral, DeepSeek, Qwen and others — are catching up to closed labs and let you run AI on your own hardware, vendor-free. That shifts the balance of power, the safety debate and the economics of the field. Here is everything on open-source AI: open-weight releases, licenses, local deployment and the race against closed models.
Meta open-sources Muse Spark Open: an 816B-parameter MoE with a 1M-token context
Meta has released Muse Spark Open, an open-weight model with 816B parameters, 42B active per token, text, image, audio and video input, and tuning for agentic…
Property Finder Found a Flaw in Its Labels and Cut VLM Costs by 86%
96.8% of Property Finder's reference labels were inherited from its old model. After relabeling, Llama 4 Maverick on Bedrock scored Micro F1 0.80 vs 0.56 at $2.35…
DeepSeek Open-Sources DeepInfer as the US and China Prepare Controls on Open Models
DeepSeek open-sourced its internal DeepInfer inference stack, with estimated gross margins up to 90%, as DHS bars it from federal procurement and MOFCOM proposes…
Open-source Hob brings Claude Cowork-style workflows to your laptop, no cloud required
Meta opens Muse Spark at 816B, Reflection releases Beam under Apache 2.0
Nepali Text "Weighed" Three Times More Than English — Tokenizer Fix Couldn't Fully Solve It
SAP researchers teach AI agents to stop flip-flopping between different solutions to the same task
DingTalk Open-Sources DWS, Brings 1,000+ Commands for AI Agents to HarmonyOS PC
DingTalk Workspace CLI now runs on HarmonyOS PC and has gone open source: 25 business modules and more than 1,000 commands let AI agents manage chats, tasks, and…
Y Combinator's Pomerado turns websites into APIs: 3 cents instead of $41
Startup Pomerado teaches AI agents to call websites directly via API/MCP instead of clicking: in a demo, a task took 8 minutes and 3 cents instead of 21 minutes and…
Artificial Analysis releases Intelligence Index v5: open models fall 12 points behind
KAIST drops calibration from MoE decoding, speeds up blocks by up to 2.69× over vLLM
Claude Opus 5.5 wins Code2Video-AA with 1536 Elo, beating GPT-6 Astra by 109
Alibaba Launches Second-Generation Qwen Glasses Starting at $450
Anthropic Reveals How It Isolates Claude: Users Simply Approved 93% of Confirmations
Anthropic detailed how it protects agents in claude.ai, Claude Code and Cowork: users approved 93% of requests, and a phishing prompt stole AWS keys in 24 of 25…
Devin now supports ChatGPT sign-in: Fusion with Astra is 39% cheaper
Devin accepts ChatGPT Plus and Pro subscriptions for OpenAI models. Running Astra + SWE-2 in Fusion mode costs 39% less than Astra alone.