Skip to content
Source · Daily Brief

AI Daily Brief — 13 February 2026

Friday’s two threads: CNBC and TechCrunch publish the Anthropic Super Bowl bump — Claude DAU +11% post-game (largest jump in their AI coverage), site visits +6.5%, and the Claude app entering Apple App Store top-10 free, beating ChatGPT, Gemini and Meta AI. Sam Altman calls the Super Bowl spots “deceptive” and “clearly dishonest.” Meanwhile OpenAI carries out the scheduled retirement of GPT-4o, GPT-4.1, GPT-4.1 mini and o4-mini from ChatGPT, with usage having shifted to GPT-5.2. Yesterday’s MiniMax M2.5 open-weights drop and ByteDance Seedance 2.0 video model continue to trend in Friday coverage.

Top stories

  • Anthropic Super Bowl bump — Claude +11% DAU. Similarweb data via CNBC shows Claude’s lift as the largest in their AI coverage, well ahead of ChatGPT (+2.7%) and Gemini (+1.4%). Site visits up 6.5%; the app climbs into App Store top-10 free, beating ChatGPT, Gemini and Meta AI. Sam Altman pushes back: the spots are “deceptive” and “clearly dishonest.” via CNBC
  • Claude breaks the App Store top 10. TechCrunch confirms the Super Bowl bump pushed the Claude app into the top 10 free apps the day after the OpenAI-bashing spot ran. via TechCrunch
  • OpenAI retires the GPT-4 family from ChatGPT. Final scheduled retirement of GPT-4o, GPT-4.1, GPT-4.1 mini and o4-mini from ChatGPT (GPT-5 Instant/Thinking already removed). Reason: only ~0.1% of users still chose GPT-4o daily; usage shifted to GPT-5.2. GPT-4o API retires Monday. via OpenAI
  • MiniMax M2.5 — community digest. Yesterday’s drop from Shanghai: 230B MoE / 10B active, open-source on Hugging Face under modified MIT (must display “MiniMax M2.5” in UI for commercial use). Benchmarks landing this week: 80.2% SWE-Bench Verified, 51.3% Multi-SWE-Bench (#1), 76.3% BrowseComp — within ~1pp of Claude Opus 4.6 and ahead of GPT-5.2 on several agentic benchmarks. Roughly $1/hour at 100 tok/s, cutting frontier cost by up to 95%. via VentureBeat

Open-source pulse

The week’s wave continues to land. Zhipu GLM-5 (Tuesday) — 744B/44B-active MoE, MIT-licensed, trained entirely on 100K Huawei Ascend 910B chips via MindSpore, 50.4% on Humanity’s Last Exam (beating Claude Opus 4.5), industry’s lowest reported hallucination rate. ByteDance Seedance 2.0 (Thursday) — multimodal text + image + audio → video, up to 15-second clips with native audio. Latam-GPT launched 10 February by Chile’s President Boric — Latin America’s first ~70B open Llama-3.1-based regional model, $550K budget. Coinbase Agentic Wallets launched 11 February for autonomous AI agents under the x402 protocol with programmable guardrails. T-Mobile‘s network-native AI Live Translation beta — 50+ languages, *87* activation, no app required.

Quiet corners

No first-party model shipping from Google DeepMind, Meta AI, xAI or Thinking Machines Lab. Anthropic takes its post-Series-G victory lap. Industry attention now turns to the India AI Impact Summit fortnight beginning 16 February.

By the numbers

  • Claude DAU post-Super Bowl: +11% (vs ChatGPT +2.7%, Gemini +1.4%) · site visits +6.5% · App Store top-10 free
  • GPT-4o ChatGPT retirement: usage share down to ~0.1% daily
  • MiniMax M2.5: 230B / 10B active · SWE-Bench Verified 80.2% · Multi-SWE-Bench 51.3% (#1) · BrowseComp 76.3% · ~95% cost cut vs frontier
  • GLM-5: 50.4% Humanity’s Last Exam · 100K Huawei Ascend 910B chips · zero NVIDIA GPUs

Compiled by AI Feed’s editor from publicly reported announcements on 13 February 2026.