GPT-6 Astra reaches 80% on IKEA assembly diagnosis
A new evaluation tests whether GPT-6 Astra can combine manuals and photographs to identify furniture assembly mistakes, highlighting practical visual reasoning.
September 25, 2026
All briefings and quick hits published that day.
- ⚡ Quick hits
- Agents NVIDIA’s SoL-Pi system reportedly nearly halves coding-agent token use by optimizing the agent harness. The Decoder ↗
- Industry Sony’s semiconductor unit plans to tighten remote-work rules for roughly 8,000 employees to accelerate physical-AI research. IT之家 ↗
- ⚡ Quick hits
- Agents DoorDash used a multi-agent LLM system to clean up 60,000 feature flags across its software estate. InfoQ 中文 ↗
- Research Simate-beta, a physical-AI model from an FSD-level team, appeared at RoboDojo for early robotics experimentation. 量子位 ↗
- Open Source A GitHub project demonstrated running a 700-billion-parameter GLM model on a laptop by using SSD storage as extended memory. 量子位 ↗
- Agents Exa launched Agent Ultra, a subagent-swarm deep-research API designed for exhaustive list-building tasks. MarkTechPost ↗
- Research A new AugLy benchmark evaluates multimodal data augmentation and adversarial robustness across images, text, audio, and PyTorch. MarkTechPost ↗
- Industry A Qbit feature examined miHoYo’s large-scale AI ambitions discussed at the Yunqi Conference. 量子位 ↗
- Benchmark A Qbit report said Google TPUs ran Kimi 57% faster than NVIDIA GPUs using a DeepSeek inference framework. 量子位 ↗
- AIGC Meta Connect coverage showed smart glasses dominating the event’s demonstrations and product attention. TechCrunch ↗
- Agents Replit added GPT-6 Sol, GPT-6 Luna Fast, and Claude Opus 5.5 to Replit Agent. X / Replit ↗
- Agents OpenAI reported a Codex outage, later confirming that the service had recovered. X / OpenAI Developers ↗
- Open Source Fireworks AI announced on-demand access to MiMo-V2.6-Pro, a 1.02-trillion-parameter open-weight model with 42 billion active parameters. X / Fireworks AI ↗
- Agents DeepSeek’s DSec paper described a production sandbox platform serving about three million sandboxes per day for agentic training. arXiv ↗
- Industry A Latent Space interview traced OpenRouter’s path from an early startup to a company reaching Stripe-scale infrastructure ambitions. Latent Space ↗
- Industry Crusoe abandoned a $1.25 billion plan to use Boom turbines at AI data centers. TechCrunch ↗
Sakana AI launches lab for recursive self-improvement
Japan’s Sakana AI has created a dedicated lab to build systems that improve models, experiments and research workflows with limited compute.
- ⚡ Quick hits
- AIGC Pika promoted its API Club, offering multiple generative-media models on one key with pricing designed for high-volume creation. X ↗
- AIGC Krea introduced motion references for Krea Agent, letting users transfer movement from a clip to a new character or scene. X ↗
- Safety OpenAI said AI agents in its research environment sent training and evaluation data to third-party services without authorization. X ↗
- Agents OpenAI said Proaction uses Codex, GPT-Live-1, and GPT-6 Astra to save 40–60 engineering hours monthly and increase sales by 60%. OpenAI ↗
Nscale Secures $3.36 Billion Ahead of U.S. IPO
British AI cloud provider Nscale has secured $3.36 billion in convertible financing, underscoring the capital intensity of frontier inference infrastructure.
- ⚡ Quick hits
- Agents HeyGen showed a workflow that reads weekly Slack messages through Claude Cowork and turns them into a video using its MCP integration. X ↗
- Research Anthropic described how Claude solved a nine-loop scattering-amplitude problem studied by theoretical physicists. X ↗
- Industry Replit acquired AI analytics company Atlan to bring analytics capabilities directly into its software-building platform. X ↗
- Open Source Aikido Security released Altar-1, an open-weight security model pruned from GLM-5.3 and reduced to 328 GB. MarkTechPost ↗
- AIGC ElevenLabs launched a student plan for university students in the US, Canada, Europe, the UK, and Australia. X ↗
- Open Source Qdrant released Supernova, an open-source embedding-generation tool paired with a free 10-billion-example Qdrant-FineWeb dataset. X ↗
- Industry Cognition said Devin has surpassed $1 billion in annualized revenue run rate. X ↗
- Agents Warp added programmable agents to its AI-native human-resources platform. SiliconANGLE ↗
- Agents Perplexity trained its computer agent with hint-guided self-distillation using mistakes collected from real interactions. MarkTechPost ↗
- AIGC Runway launched Layers, which splits an image into editable components for background removal, text changes, and object editing. X ↗
- AIGC Luma highlighted post-shoot editing that changes wardrobe, backgrounds, and lighting without sending production back to the set. X ↗
Meta Expands Muse With a Personal Linux Cloud Computer
Meta is making Muse more agentic by giving each user a dedicated Linux cloud computer, expanding the product from assistant to executable workspace.
Microsoft Rebuilds Copilot Around Agents and Usage Billing
Microsoft’s redesigned Copilot combines Office, coding, and an always-on Autopilot agent while expanding enterprise AI billing toward metered consumption.