Skip to content
Source · Daily Brief

AI Daily Brief — 16 December 2025

Tuesday packed three structural launches. OpenAI shipped GPT Image 1.5 — new ChatGPT Images model with stronger instruction-following, more precise edits preserving untouched elements, up to 4× faster generation. Rolling out to all ChatGPT users + API; API I/O 20% cheaper than GPT Image 1. Framed as part of Sam Altman’s “Code Red” push against Google’s Gemini. OpenAI published FrontierScience benchmark for PhD-level scientific reasoning — physics + chemistry + biology with 2 tracks: 100 Olympiad-style constrained problems + 60 open-ended Research tasks, validated by international science medalists + PhD scientists. GPT-5.2 leads: 77% Olympiad + 25% Research, ahead of Claude Opus 4.5 + Gemini 3 Pro. Results show stark gap between closed-form scientific reasoning + open-ended research: top model just 25% on Research vs 77% Olympiad. Evaluated against Claude Opus 4.5 + Gemini 3 Pro + GPT-4o + o4-mini + o3. Meta begins global rollout of v21 update to Ray-Ban Meta + Oakley Meta HSTN glasses. New “Conversation Focus” uses 5-microphone beamforming array + Snapdragon AR1 Gen1 to amplify voice of whoever wearer is looking at, suppressing background noise with sub-10ms latency. Also added: multimodal Spotify integration (“play a song to match this view”), Hyperlapse up to 30 minutes, slow-motion capture, Find Device, Garmin/Strava voice fitness controls. Early access Conversation Focus US + Canada. Microsoft Foundry adds GPT Image 1.5 to Azure AI Foundry day-one — demonstration of still-tight OpenAI/Microsoft model-release pipeline despite Microsoft’s growing in-house MAI effort. SiliconANGLE: GPT Image 1.5 OpenAI’s most capable text-to-image — editing-first; preserves unchanged elements (lighting + composition + faces) across multi-turn edits; improved dense text rendering. Android Central: Ray-Ban Meta v21 holiday listening update — most significant Ray-Ban Meta software update since launch.

Top stories

  • OpenAI ships GPT Image 1.5. Stronger instruction-following + precise edits + 4× faster + API I/O 20% cheaper than GPT Image 1. Free across ChatGPT + API. via OpenAI
  • OpenAI publishes FrontierScience benchmark — PhD-level physics + chem + bio. 100 Olympiad + 60 open-ended Research tasks. PhD + medalist validated. GPT-5.2 leads 77% Olympiad / 25% Research. via OpenAI
  • OpenAI X: FrontierScience announcement. “Hard expert-written questions (olympiad + longer)” measuring PhD-level reasoning. via OpenAI / X
  • TechCrunch: GPT Image 1.5 “Code Red” counter to Google. Continuing warpath in response to Gemini momentum. via TechCrunch
  • Meta v21 update to Ray-Ban + Oakley Meta glasses globally. Conversation Focus (5-mic beamforming + Snapdragon AR1 Gen1, sub-10ms latency). Multimodal Spotify scene-matching. Hyperlapse 30 min. Slow-mo. Find Device. Garmin/Strava voice fitness. via Meta
  • Meta Quest blog: v21 full release notes. Conversation Focus US + Canada early access + multimodal “soundtrack-your-world” Spotify. via Meta Quest
  • Microsoft Foundry adds GPT Image 1.5 day-one. Demonstrates still-tight OpenAI/Microsoft model-release pipeline despite Microsoft’s in-house MAI. via Microsoft
  • SiliconANGLE: GPT Image 1.5 most-capable text-to-image OpenAI. Editing-first; preserves unchanged elements across multi-turn edits; improved dense text rendering. via SiliconANGLE
  • FrontierScience: frontier models weak on open-ended research. Top model just 25% on 60-task Research vs 77% Olympiad. Claude Opus 4.5 + Gemini 3 Pro + GPT-4o + o4-mini + o3 all evaluated. via TIME
  • Android Central: Ray-Ban Meta v21 holiday update. Most significant Ray-Ban Meta software update since launch. Spotify scene-based playlists + conversation amplification. via Android Central

Who shipped

OpenAI shipped GPT Image 1.5 + FrontierScience benchmark. Meta shipped Ray-Ban + Oakley v21. Microsoft shipped Foundry GPT Image 1.5 day-one. Anthropic, Google, xAI, DeepSeek, Alibaba quiet on direct launches.

Open-source pulse

No new Chinese drops. NVIDIA Nemotron 3 family from Mon continues community uptake.

Money, infra & hardware

Meta v21 + Conversation Focus = first major consumer-AI hardware update with multimodal scene-matching + sub-10ms latency claim. Microsoft Foundry day-one GPT Image 1.5 signals continued OpenAI distribution alignment.

Quiet corners

FrontierScience benchmark gap (77% Olympiad vs 25% open-ended Research) is the substantive evaluation moment — measures real research capability vs constrained problem-solving.

By the numbers

  • 4× / 20% — GPT Image 1.5 speed / API cost reduction
  • 100 / 60 / 77% / 25% — FrontierScience Olympiad tasks / Research tasks / GPT-5.2 Olympiad / Research
  • 5-mic / sub-10ms / 30 min — Meta v21 beamforming array / Conversation Focus latency / Hyperlapse max
  • Most-mentioned company: OpenAI
  • Quietest segment: Chinese launches

Compiled by AI Feed’s editor from verified web sources for 16 December 2025.