Skip to content
Source · Daily Brief

AI Daily Brief — 17 December 2025

Wednesday compressed two structural shifts. Google launches Gemini 3 Flash — pairs Pro-grade reasoning with Flash-level latency. Replaces Gemini 2.5 Flash as default in Gemini app + rolls out as default for AI Mode in Search globally. SWE-bench Verified 78% (beats Gemini 3 Pro on coding); MMMU-Pro 81.2%; GPQA Diamond 90.4%; HLE 33.7% no-tools. 1M context + configurable thinking levels (minimal/low/medium/high). $0.50/$3 per Mtoken ($1/M audio in) — ~67% above 2.5 Flash but materially stronger reasoning. Available in Gemini API (AI Studio) + Antigravity + Gemini CLI + Android Studio + Vertex AI + Gemini Enterprise. Launch customers: JetBrains + Figma + Cursor + Harvey + Latitude. Beat Gemini 2.5 Pro on most benchmarks while ~3× faster on Artificial Analysis. Nano Banana Pro available in AI Mode alongside for image gen. Mythic raises oversubscribed $125M led by DCVC to challenge NVIDIA with analog AI chips — NEA + Atreides + Future Ventures + SoftBank SBVA + Linse Capital + S3 Ventures + Honda + Lockheed Martin. Pitches 100× energy efficiency over GPUs/ASICs across edge (robotics + automotive + defense) + 1T+ parameter LLMs in data centers. Total funding past $175M. SiliconANGLE: compute-in-memory architecture aims at AI’s power wall. Mythic: “GPU pantheon” challenge with analog chiplets. Apple confirms major Siri AI upgrade slips into 2026 — CNBC: exec said Apple “didn’t want to disappoint customers.” Characterizes 2025 as the year Apple punted on AI while OpenAI shipped Sora 2 + Anthropic shipped multiple Claudes + Amazon revamped Alexa. European Commission publishes first draft Code of Practice on AI-generated content labeling — voluntary Code implementing Article 50 of EU AI Act covering watermarking + detection of synthetic media by providers + deployer obligations to label deepfakes + AI-generated text on matters of public interest. Sets technical groundwork ahead of binding transparency rules on Aug 2, 2026. Antigravity adds Gemini 3 Flash for agentic coding — “agentic coding at the speed of thought.”

Top stories

  • Google launches Gemini 3 Flash — default Gemini app + AI Mode in Search global. Pro-grade reasoning + Flash latency. via TechCrunch
  • Gemini 3 Flash benchmarks: SWE-bench Verified 78% / MMMU-Pro 81.2% / GPQA Diamond 90.4% / HLE 33.7%. Beats Gemini 2.5 Pro on most; even outpaces Gemini 3 Pro on SWE-bench. 1M context. via Google
  • Gemini 3 Flash $0.50/$3 per Mtoken; $1/M audio. Across Gemini API + Antigravity + Gemini CLI + Android Studio + Vertex + Gemini Enterprise. Launch customers: JetBrains + Figma + Cursor + Harvey + Latitude. via Google Developers
  • Mythic $125M oversubscribed for analog AI chips (DCVC lead). NEA + Atreides + Future Ventures + SoftBank SBVA + Linse + S3 + Honda + Lockheed Martin. 100× energy efficiency over GPUs/ASICs (edge + 1T+ LLMs). Total >$175M. via Bloomberg
  • SiliconANGLE: Mythic compute-in-memory at AI power wall. Analog chiplets bet on breaking AI energy bottleneck edge + DC. via SiliconANGLE
  • Apple confirms major Siri AI upgrade slips to 2026. CNBC: “didn’t want to disappoint customers.” Frames 2025 as Apple’s lost AI year vs Sora 2 + multiple Claudes + Alexa revamp. via CNBC
  • EU draft Code of Practice on AI-generated content labeling. Implements Article 50 EU AI Act. Watermarking + synthetic-media detection + deployer deepfake labeling. Sets groundwork for Aug 2, 2026 binding transparency rules. via European Commission
  • Antigravity adds Gemini 3 Flash for agentic coding. “Agentic coding at the speed of thought.” Pro-class reasoning at Flash latency. via Antigravity
  • Gemini 3 Flash in AI Mode for Search + Nano Banana Pro in AI Mode. Default query reasoning model globally. via Google
  • Mythic IR release: GPU pantheon challenge. Oversubscribed $125M. Robotics + automotive + defense + DC LLMs. via Mythic
  • 9to5Google: Gemini 3 Flash “Pro-level performance” ~3× faster. Matches/beats Gemini 2.5 Pro on most benchmarks per Artificial Analysis. via 9to5Google

Who shipped

Google shipped Gemini 3 Flash + AI Mode default + Antigravity integration. Mythic shipped $125M round + analog AI silicon narrative. EU Commission shipped draft Code of Practice. Apple shipped Siri delay confirmation. OpenAI, Anthropic, Meta, xAI quiet on launches.

Open-source pulse

No fresh open-source drops Dec 17. NVIDIA Nemotron 3 family from Mon continues HF uptake. Chinese 7-of-10 dominance continues.

Money, infra & hardware

Mythic $125M = first major analog AI chip funding event at scale. 100× energy efficiency claim against AI’s energy bottleneck. Combined with NVIDIA Nemotron 3 + Trainium3 + d-Matrix + Cerebras + Extropic TSU = ~6 distinct AI silicon programs racing.

Quiet corners

EU Code of Practice draft is the substantive AI Act enforcement build-up moment — sets technical floor for Article 50 transparency requirements binding Aug 2, 2026. Apple’s punt on Siri AI consolidates “Apple lost 2025” narrative.

By the numbers

  • 78% / 81.2% / 90.4% / 33.7% — Gemini 3 Flash SWE-Verified / MMMU-Pro / GPQA Diamond / HLE no-tools
  • $0.50 / $3 / 3× — Gemini 3 Flash input / output per Mtoken / speed vs 2.5 Flash
  • $125M / 100× / $175M+ — Mythic round / energy efficiency claim / total funding
  • 5 / Aug 2, 2026 — Gemini 3 Flash launch customers (JetBrains + Figma + Cursor + Harvey + Latitude) / EU AI Act binding transparency date
  • Most-mentioned company: Google
  • Quietest segment: Anthropic launches

Compiled by AI Feed’s editor from verified web sources for 17 December 2025.