Skip to content
Source · Daily Brief

AI Daily Brief — 19 February 2026

A heavyweight Thursday. Google ships Gemini 3.1 Pro in preview — the first .1 increment in the Gemini cadence — with 77.1% on ARC-AGI-2 (more than double Gemini 3 Pro’s 31.1%), 80.6% on SWE-Bench Verified, and 94.3% on GPQA Diamond. It leads 13 of 16 benchmarks tested. Pricing holds flat at $2 / $12 per M input/output ($4/$18 above 200K tokens). The model rolls out same-day across the Gemini app, NotebookLM (AI Pro and Ultra), the Gemini API in AI Studio, Vertex AI, Gemini Enterprise, the Gemini CLI, Android Studio and Google Antigravity. In New Delhi, PM Modi delivers the inaugural address at the India AI Impact Summit; 13 frontier labs sign the “New Delhi Frontier AI Impact Commitments”; Tata and OpenAI announce a multi-dimensional strategic partnership; and Pichai, Altman, Hassabis, Amodei and Brad Smith are all on the same stage.

Top stories

  • Google — Gemini 3.1 Pro. Reasoning-focused preview model; 1M-token context across text, images, audio, video, PDFs and entire code repositories. ARC-AGI-2 77.1%, SWE-Bench Verified 80.6%, GPQA Diamond 94.3%, leading 13 of 16 benchmarks. via Google
  • Google Antigravity gets 3.1 Pro for agentic coding. Positioned as “optimised for software engineering behaviour and agentic workflows that require precise tool usage and reliable multi-step execution.” Antigravity gives agents dedicated surfaces with direct access to editor, terminal and browser to plan and execute end-to-end software tasks while self-validating code. via Google Cloud
  • Modi inaugurates the India AI Impact Summit. Modi calls for AI to be “democratic, inclusive and guided by strong ethical principles,” unveils India’s M.A.N.A.V. framework (Moral/Ethical, Accountable Governance, National Sovereignty, Accessible/Inclusive, Valid/Legitimate), and issues the call “Design and Develop in India, Deliver to the World, Deliver to Humanity.” Opening ceremony co-addressed by French President Emmanuel Macron and UN Secretary-General Antonio Guterres. via PMO India
  • 13 frontier labs sign the New Delhi Frontier AI Impact Commitments. Two commitments: publish anonymised aggregated insights into how their AI systems are used in the real world (jobs, skills, productivity); strengthen testing and evaluation of frontier models on underrepresented languages and Global South cultural contexts. Pitched as an inclusion-and-transparency complement to the Bletchley/Seoul/Paris safety commitments line. via PIB
  • Tata × OpenAI — multi-dimensional strategic partnership. Covers AI-led innovation across Tata Group companies, joint enterprise AI transformation worldwide, and AI infrastructure build-out in India. Sam Altman keynotes that OpenAI will open Bengaluru + Mumbai offices later in 2026 (adding to New Delhi). via News India Times
  • OpenAI academic initiative — first Indian cohort. IIT Delhi, IIM Ahmedabad, Manipal Academy of Higher Education and AIIMS join an OpenAI academic programme expected to support more than 100,000 students, faculty and staff in the first year. via Digit

Frontier-lab convocation

One stage in New Delhi seats Sundar Pichai (Alphabet/Google), Sam Altman (OpenAI), Demis Hassabis (Google DeepMind), Dario Amodei (Anthropic) and Brad Smith (Microsoft) alongside 20+ heads of state and 60+ ministers from 100+ countries — a notable concentration of frontier-lab leadership at a non-Western event.

Quiet corners

No first-party shipping from Anthropic, Meta AI, xAI, Microsoft AI or Thinking Machines Lab. DeepSeek stays silent through the rumoured post-LNY window.

By the numbers

  • Gemini 3.1 Pro: ARC-AGI-2 77.1% (2.5× Gemini 3 Pro) · SWE-Bench Verified 80.6% · GPQA Diamond 94.3% · leads 13/16 benchmarks tested
  • Gemini 3.1 Pro pricing: $2 / $12 per M tokens ($4 / $18 above 200K) · 1M-token context
  • New Delhi Frontier AI Impact Commitments: 13 signatory labs
  • India AI Impact Summit: 20+ heads of state · 60+ ministers · 100+ countries
  • OpenAI India academic initiative: 100,000+ students/faculty/staff in year one

Compiled by AI Feed’s editor from publicly reported announcements on 19 February 2026.