AI News: India AI Summit concludes, Claude Sonnet 4.6 goes default, Saudi backs xAI with $3B

Daily roundup for February 20, 2026 covering the India AI Summit wrap-up, Claude Sonnet 4.6 becoming the default model, Saudi Arabia's $3 billion xAI investment, OpenAI's India education push, UK deepfake laws, and Google I/O 2026 dates.

Top Stories

India AI Summit Concludes: Global South Claims Seat at AI Governance Table

The India AI Impact Summit wrapped today in New Delhi after five days of negotiations, announcements, and the usual diplomatic theater. The event drew delegations from over 100 countries, including more than 20 heads of state, with French President Emmanuel Macron and UN Secretary-General António Guterres addressing the opening ceremony alongside PM Modi.

The concrete outcomes: several bilateral AI cooperation agreements, India’s commitment to expand its IndiaAI Mission compute infrastructure, and a joint declaration on “inclusive AI development” that 87 nations signed. Whether these translate to meaningful coordination or join the pile of unimplemented international AI declarations remains to be seen.

The subtext matters more. By hosting the first major AI summit outside Western capitals or Beijing, India positioned the Global South as a potential third pole in AI governance. China skipped the summit; the U.S. sent a delegation but no cabinet-level officials. The vacuum creates opportunity - or just fragmentation, depending on whether anyone fills it with actual coordination mechanisms.

Sources: India AI Summit, PMIndia

Claude Sonnet 4.6 Now Default: Opus-Level Performance at Sonnet Prices

Anthropic’s Claude Sonnet 4.6 became the default model for Free and Pro users this week, just three days after release. The model achieves 72.5% on OSWorld-Verified - 0.2% behind Opus 4.6 - while costing one-fifth as much ($3/$15 per million tokens vs $15/$75).

Early access developers preferred Sonnet 4.6 over its predecessor 70% of the time for coding tasks. More surprising: 59% preferred it over Opus 4.5. The model includes a 1M token context window (in beta), extended thinking support, and significantly improved computer use capabilities.

For the AI industry, this continues the pattern where mid-tier models rapidly close the gap with flagships. If Sonnet 4.6 genuinely matches Opus 4.6 at 20% of the cost, the pricing pressure on frontier models intensifies. For users, it means near-flagship capability became free today.

Sources: Anthropic, CNBC, VentureBeat

Saudi Arabia’s Humain Invests $3 Billion in Musk’s xAI

Saudi Arabian AI company Humain took a $3 billion stake in Elon Musk’s xAI as part of the company’s $20 billion funding round - completed just before xAI’s acquisition by SpaceX. Humain becomes a “significant minority shareholder,” with holdings converting to roughly 0.24% of the combined $1.25 trillion SpaceX entity.

The investment builds on a November 2025 partnership where Humain and xAI committed to developing 500+ megawatts of AI data center infrastructure in Saudi Arabia and deploying xAI’s Grok models in the Kingdom.

The deal deepens ties between Musk and Saudi Arabia’s Public Investment Fund, which created Humain in May 2025 as its national AI champion. For Saudi Arabia, it’s another move in its aggressive AI strategy. For xAI, it’s validation and capital - though integrating into SpaceX while building Saudi infrastructure creates an interesting governance complexity.

Sources: Bloomberg, CNBC

Quick Hits

  • OpenAI expands into Indian higher education: Partnering with six institutions including IIT Delhi, IIM Ahmedabad, and AIIMS to provide ChatGPT Edu access, faculty training, and certifications to 100,000+ students. Also working with ed-tech platforms Physics Wallah and upGrad. The India push coincides with the AI Summit and OpenAI’s broader emerging-market strategy. TechCrunch

  • UK mandates 48-hour takedown for deepfakes: PM Starmer announced amendments to the Crime and Policing Bill requiring platforms to remove non-consensual intimate images within 48 hours. Creating such images became illegal on February 6. Nudification apps are now explicitly outlawed. The 48-hour window applies to AI-generated and photoshopped content alike - this is treated as seriously as terrorism content. The Register, GOV.UK

  • Google I/O 2026 confirmed for May 19-20: The annual developer conference returns to Shoreline Amphitheatre in Mountain View. Expect Gemini updates, Android announcements, and whatever else Google has cooking. Sundar Pichai announced the dates while attending the India AI Summit. 9to5Google

  • Apple’s March 4 event: Apple announced a “special Apple Experience” in New York, London, and Shanghai. Expected: iPhone 17e, MacBook Pro with M5 chips, new iPads. Notably, Apple said “experience” rather than “event” - possibly signaling a different format. 9to5Mac

  • Meta Reality Labs cuts continue: The 1,500 layoffs (10% of Reality Labs) completed this month shift focus from VR to wearables and AI. Horizon Workrooms shut down. Reality Labs has lost over $80 billion since 2020. Ray-Ban Meta glasses and similar wearables are now the priority. GeekWire

  • DeepSeek V4 expected imminently: The Chinese lab’s next major model reportedly targets mid-February launch around Lunar New Year. Claimed specs: 1 trillion parameters, 1M+ context window, 80%+ SWE-bench at 10-40x lower inference costs than Western competitors. Open-weight release expected. Introl

Worth Watching

The pricing dynamics in the model market are accelerating. Sonnet 4.6 matching Opus at 20% cost, Chinese models claiming GPT-5.2 parity at 60% lower cost, DeepSeek promising 80%+ SWE-bench at 10x lower cost - if even half these claims hold up, the economics of AI deployment shift dramatically. Enterprise buyers should be questioning any long-term commitments at current API pricing.

The UK’s deepfake law is notable for treating AI-generated intimate images equivalently to terrorism content in terms of platform obligations. That’s a regulatory template other countries may follow. The 48-hour enforcement window will pressure platforms toward more aggressive automated detection - with the usual concerns about false positives and appeals processes.

Meta’s Reality Labs pivot is now complete: $80 billion spent on the metaverse vision, Horizon Workrooms abandoned, VR studios closed. The company’s future is AI and wearables, not virtual worlds. That’s a significant strategic admission, even if the official messaging frames it as “evolution.”