Grok 4.6's independent numbers arrive — top-three intelligence at a third of the price
When xAI shipped Grok 4.6 on 7 August it came with no model card and no benchmarks, so we said to wait for third-party numbers. Those numbers are now in. On the Artificial Analysis Intelligence Index, Grok 4.6 scores 61 — tied with OpenAI's GPT-5.6 Sol Max for third, behind Anthropic's Claude Opus 5 (63) and Fable 5 Max (62), the two models above it. That ranks it third overall and first among non-Anthropic models on the index. On the GDPval-AA v2 work-task Elo it scores 1,753, which Artificial Analysis places second — behind only Claude Opus 5 and statistically level with Fable 5.
The pricing is where it bites. Grok 4.6 lists at $2 per million input tokens and $6 per million output, against Sol's $5 and $30 (the-decoder) — roughly a third of the cost for comparable index scores, and a fifth of the price on output. For an EU buyer the performance-per-euro is real, but the model is a closed, US-hosted API: the cost gap does not change which law reaches the data you send it.
Lovable raises $400M at $13.3B — Europe's most valuable AI app-builder
Stockholm-based Lovable raised a $400M Series C at a $13.3 billion valuation, Bloomberg reported on 12 August — double its $6.6 billion mark from December. The round was led by Menlo Ventures and the Scaleup Europe Fund, an EU vehicle run by EQT. Lovable lets people build web apps by describing them in plain language; Yahoo Finance notes annual recurring revenue has nearly tripled toward a $600M run rate, with Nvidia, Deutsche Telekom and Adidas among those building on it.
The raise makes Lovable one of Europe's most valuable private tech companies and a rare European name in a “vibe-coding” field otherwise dominated by US players. For regulated buyers the interesting question is not the valuation but the deployment model: where the generated app runs, and whether prompts and business data leave the building — the same line AWS drew when it moved Superblocks builds inside customers' own cloud earlier this month.
Anthropic will watermark all Claude text worldwide
Anthropic said on 11 August it will embed machine-readable watermarks in text from new Claude models and attach C2PA-signed provenance metadata to generated image files, TechCrunch reported. It is a direct response to the EU AI Act's Article 50 transparency rules, which became enforceable on 2 August — but Anthropic is applying the marking globally, not just to EU users, a clear case of one jurisdiction's rule becoming a vendor's default everywhere (Euronews).
The limits matter for anyone treating this as a compliance control. A detected mark shows Claude likely processed the content, not that it authored it, and file metadata can be stripped by format conversion, re-saving or a screenshot. Provenance labelling is now a procurement checkbox for EU deployers; it is not yet a guarantee.
Quick Hits
- Largest AI supply-chain breach of 2026 disclosed — CloudSEK detailed a campaign, newly disclosed on 11 August but dating to a March compromise of LiteLLM PyPI packages, that exposed 2,500+ companies and roughly 434,000 CI/CD pipelines, with named exposure spanning Nvidia, AWS and Cisco (CloudSEK). The AI plumbing layer is now a first-class attack surface.
- AI took ~86% of US venture dollars — PitchBook data cited by Fortune shows AI captured roughly 86% of US VC in the first half of 2026, with $100M-plus megadeals soaking up 87.5% of all dollars — concentration on both axes.
- Skan AI raised $63M Series C — the San Jose process-intelligence firm's round was co-led by Cathay Innovation and Dell Technologies Capital, with Citi Ventures and Bloomberg Beta joining (FinSMEs).