Source
· Daily Brief
AI Daily Brief — 3 April 2026
A digestion Friday — the news cycle absorbs Wednesday’s Gemma 4 drop. Early benchmark numbers land for Google’s open family: the 31B Dense scores 89.2% on AIME 2026, 80.0% on LiveCodeBench v6, 86.4% on Agentic Tool Use, and a 20× Codeforces ELO jump over Gemma 3.
Top stories
- Gemma 4 benchmarks land. The 31B Dense secures a top-three global ranking on the Arena AI leaderboard among open models — outperforming closed models nearly four times its size. The 31B carries 256K context, all four variants handle 140+ languages, and the licence is clean Apache 2.0 with no custom clauses. via Labellerr
- Native video for every variant. All four Gemma 4 models — 2B, 4B, 27B MoE and 31B Dense — ship with native video understanding, with native audio input on the E2B and E4B variants. via Google AI for Developers
Quiet corners
No new frontier model dropped on 3 April. The week’s flagship table holds — GPT-5.4 Thinking, Claude Sonnet 4.6, Gemini 3.1 Pro, Grok 4.20 Beta 2 — with attention now on what Anthropic and OpenAI counter with next.
By the numbers
- 89.2% Gemma 4 31B on AIME 2026 (math)
- 80.0% on LiveCodeBench v6
- 256K context window on the 31B Dense
- 140+ languages supported across all variants
Compiled by AI Feed’s editor from publicly reported announcements on 3 April 2026.