Skip to content
Source · Daily Brief

AI Morning Brief — 13 August 2026

SpaceXAI launched Grok 4.6 overnight, reaching the top of the GDPVal-AA benchmark and matching GPT-5.6 Sol on the Artificial Analysis Intelligence Index at lower cost. Alibaba simultaneously released Qwen3.8-2.4T-A95B — a 2.4-trillion-parameter open-weight MoE — and DeepSeek quietly pushed V4 Pro 0813 to API with no official announcement page.

Top stories

  • Grok 4.6 and Grok Bot launch. SpaceXAI released Grok 4.6, scoring 61 on the Artificial Analysis Intelligence Index (tied with GPT-5.6 Sol), reaching #1 on GDPVal-AA at 1,753 ELO, and tying Claude Opus 5 Max on the Agentic Index — at a price more than 60 percent below comparable frontier models. Alongside it, SpaceXAI launched Grok Bot: persistent AI teammate agents that can log into apps, complete multi-step tasks autonomously, and surface results asynchronously. via THE DECODER
  • Qwen3.8-2.4T-A95B released. Alibaba open-sourced Qwen3.8-Max, its largest model at 2.4 trillion parameters with 95B active. NVIDIA demonstrated serving it on GB300 NVL72 and Together AI offered day-zero API access. via Hugging Face
  • DeepSeek V4 Pro 0813 quietly released. DeepSeek pushed a new Pro checkpoint to API via OpenRouter with no public announcement page; early benchmarks show it beating Nemotron Ultra on agentic tasks. Open weights availability remains unconfirmed. via Simon Willison
  • Anthropic watermarking draws user backlash. TechCrunch reported widespread complaints from Claude users about a new output watermarking system they say will expose their AI use at work and in academic settings. via TechCrunch
  • Supply chain attack leaks terabytes of credentials. A compromised AI Python package was used to scrape and exfiltrate credentials from roughly 2,500 users; the scope makes it one of the larger AI-ecosystem security incidents on record. via Ars Technica
  • Lovable confirms $400M Series C at $13.3B valuation. The AI app-building platform disclosed the raise after crossing $500M annualized revenue in June; apps on the platform now receive 900M visits per month. via TechCrunch

Who shipped

SpaceXAI dominated the cycle with Grok 4.6’s frontier-tier benchmark scores and the debut of Grok Bot as a persistent agentic product. Alibaba Qwen shipped Qwen3.8-Max, its largest open-weight release yet at 2.4 trillion parameters. Microsoft announced MAI-Thinking-1, described as the company’s first reasoning model built from scratch, now available in Microsoft Foundry. Sakana AI updated Sakana Chat with the new Fugu orchestrator, an updated Namazu Japanese LLM, and free code execution requiring no login.

Open-source pulse

Qwen3.8-2.4T-A95B is the headline open-source drop of the cycle — Alibaba’s largest open-weight model at near-frontier scale, with FP8 weights on Hugging Face. Liquid AI released LFM2.5-VL-3B, a compact 3B multimodal model targeting on-device and mobile inference, demonstrated running locally on an iPhone 17.

Money, infra & hardware

Lovable confirmed $400M at a $13.3B valuation. OpenAI-backed Thrive Holdings closed $2B at $12B from SoftBank, D1 Capital Partners, and Altimeter. Cognition is reportedly in talks to raise at a $40B valuation, up from $26B earlier in 2026. On hardware, NVIDIA raised the RTX PRO 6000 list price to $16,000, nearly double its original price.

Quiet corners

Meta posted no primary-source AI content in the window; community benchmarking of Muse Glimmer continued without any announcement. Mistral announced EU-resident data routing and priority queue access — both behind a surcharge — but released no new model.

By the numbers

  • 211 stories in 24 h across 38 sources
  • Most-mentioned model: Grok
  • Most-mentioned lab: SpaceXAI
  • Notable absences: Meta, Mistral

Compiled by AI Feed’s editor from all 211 headlines published on 13 August 2026.