OpenAI and Anthropic are in a full price war across model tiers, while a 31-day EU deepfake labeling deadline and India's second AI unicorn in weeks quietly reshaped the global landscape.
Anthropic's Claude Sonnet 5 is back globally, and it's now the default
Anthropic released Claude Sonnet 5 on June 30, but it was briefly blocked by a US export-control order. It returned globally on July 1 and is now the new default model on Claude.ai. It sits between the older Sonnet and the flagship Opus 4.8 in capability, but it behaves more like an AI agent, it can browse the web, run code in a terminal, and complete multi-step tasks on its own. Introductory pricing is $2 per million input tokens through August 31.
Why it matters: The model you get when you open Claude is now significantly more capable and more autonomous than it was a week ago. If you use Claude for work, the experience has quietly upgraded without you doing anything.
OpenAI previews GPT-5.6: three models, one big price cut
OpenAI previewed GPT-5.6, a three-model family: Sol (top-tier flagship), Terra (comparable quality to GPT-5.5 at roughly half the cost), and Luna (fast and very affordable for everyday tasks). Sol includes a new "ultra" mode that delegates tasks to subagents automatically. The family is in limited preview now with broader rollout expected shortly. Terra is positioned as the workhorse most users will reach for.
Why it matters: Advanced AI is getting significantly cheaper. Terra and Luna make ChatGPT-level quality accessible at a fraction of previous costs, good news for students, freelancers, and small businesses watching their spend.
EU deepfake labeling becomes law on August 2, companies are scrambling
The EU AI Act's Article 50 transparency obligations kick in on August 2, 2026, just 31 days away. From that date, any AI-generated image, video, audio, or text must be marked in a machine-readable format so it can be identified as artificially made. Deepfake rules apply even when there is no intent to deceive, if it looks or sounds like a real person, it must be labeled. Non-compliance carries fines of up to €7.5 million or 1.5% of global revenue.
Why it matters: This is the first major law that requires AI-generated content to identify itself automatically. Every app, platform, or tool that creates synthetic media (from marketing to entertainment) needs to be compliant by next month. India-facing platforms with EU users are not exempt.
Sarvam AI raises $234M and joins India's fast-growing unicorn club
Sarvam AI, the Bengaluru-based startup building a full-stack sovereign AI system for India, completed the first close of a $300M Series B round, raising $234M and reaching a $1.5 billion valuation. It joins Neysa (which raised $600M earlier this year) as India's newest AI unicorns. Indian AI startups as a whole raised $3.94 billion in Q1 2026 alone, more than six times the total for all of 2025. The government's IndiaAI Mission now has 38,000 GPUs provisioned for startups at below-market rates.
Why it matters: India is no longer just a consumer of global AI tools, it is actively building its own. Sarvam's focus on Indian languages and local use cases means AI that actually works for 1.4 billion people, not just English speakers.
1 in 3 job applicants is now an AI deepfake. HR has no good answer yet.
Pindrop Security analysed 300 job applicant profiles and found that over one-third were entirely fabricated: AI-generated resumes, fake work histories, and deepfake video interviews that can pass in real time. Gartner projects that 1 in 4 global job candidate profiles will be fake by 2028. Several US states now require employers to disclose AI use in hiring, but no law yet requires detecting AI use by applicants.
Why it matters: If you're job hunting, know that the bar for proof of authenticity is rising. If you're hiring, the screening process you used six months ago may not catch what's entering your pipeline today.
Anthropic in talks with Microsoft to run Claude on custom Maia 200 chips
Anthropic is in early-stage discussions with Microsoft to run Claude inference workloads on Microsoft's Maia 200 AI chips via Azure. The Maia 200 chips, launched in January 2026, claim over 30% better performance per dollar compared to rival silicon. The deal would make Claude available through Azure's enterprise cloud, potentially reaching millions of business users who live inside Microsoft's ecosystem.
Why it matters: The AI chip war is starting to matter for everyday users. When models run on cheaper, faster chips, prices drop and response times improve. If this deal closes, Claude gets faster and potentially cheaper for everyone using it through Microsoft tools.
The EU AI Act's August 2 countdown, and whether it actually has teeth
In 31 days, every company producing AI-generated content for EU users must label it in a machine-readable format or face fines up to €7.5M. This is the first major AI transparency law to go live anywhere. The open question is enforcement: the EU has the rules but enforcement capacity across 27 member states is untested at this scale. Whether August 2 marks a real shift in how AI content is labeled (or becomes another compliance checkbox that gets ignored) will tell us a lot about whether global AI regulation has any real grip.
AI speech transcription upgrades + EdTech funding keeps climbing
Willow.ai launched Atlas-1, a high-end speech-to-text model built for sales calls, meetings, and compliance records, a direct push into a space dominated by older tools. Separately, AI edtech startup Gizmo raised $22M in a Series A to make studying "addictive" using AI-generated quizzes and spaced repetition. Language learning apps (Preply, Speak, ELSA and others) collectively raised over $400M since 2023, making it the best-funded sub-sector in AI education.