Skip to content
Source · Daily Brief

AI Daily Brief — 31 January 2025

Friday closed January with OpenAI’s direct answer to the DeepSeek shock. o3-mini and o3-mini-high shipped in ChatGPT and the API, including — for the first time — to free-tier ChatGPT users via a “Reason” button. API pricing landed at $1.10 per million input tokens and $4.40 per million output tokens, with a 200K context window and three reasoning-effort settings. Benchmarks: 87.3% AIME 2024, 2130 Codeforces Elo, 49.3% SWE-bench Verified — OpenAI’s highest released SWE-bench Verified score at the time. Sam Altman held a Reddit AMA conceding OpenAI has been “on the wrong side of history” regarding open source. Italy’s Garante formally blocked DeepSeek; Ireland and Belgium opened parallel probes. Trump’s EO 14179 published in the Federal Register. OpenAI updated Stargate with an RFP for land and power and an RFQ for A/E firms.

Top stories

  • OpenAI releases o3-mini in ChatGPT and the API — first reasoning model on the free tier. Plus, Team, and Pro users got access on launch day (Enterprise February); free users access o3-mini via a “Reason” button or by regenerating a response. Plus/Team rate limits tripled to 150 messages/day; Pro unlimited. Function calling, Structured Outputs, and developer messages supported. Three reasoning-effort settings: low/medium/high. via OpenAI · via MIT Tech Review
  • o3-mini benchmarks: 87.3% AIME 2024, 2130 Codeforces, 49.3% SWE-bench Verified. With high reasoning effort, o3-mini hit OpenAI’s highest released SWE-bench Verified score at the time. The reasoning-effort dial swings accuracy by 10-30% from low to high across AIME, GPQA Diamond, and Codeforces. API priced $1.10 input / $4.40 output per million tokens — a direct response to DeepSeek R1 pricing. via Artificial Analysis
  • Sam Altman Reddit AMA: OpenAI “on the wrong side of history” on open source. Altman with Mark Chen, Kevin Weil, Srinivas Narayanan, Michelle Pokrass, and Hongyu Ren. Called DeepSeek “a really great model,” conceded OpenAI’s lead would narrow, said full o3 will arrive “more than a few weeks, less than a few months,” and signaled a new open-source strategy. No GPT-5 timeline. via TechCrunch
  • Italy’s Garante blocks DeepSeek; Ireland and Belgium open parallel probes. Hangzhou and Beijing DeepSeek entities told regulators they don’t fall under EU data-privacy laws; Garante ordered the app removed from Italian stores and prohibited Italian-user data processing. The first formal full block by a Western regulator. via Euronews
  • EO 14179 published in the Federal Register. Trump’s “Removing Barriers to American Leadership in AI” landed as document 2025-02172 (90 FR 8741), formalizing the 180-day AI Action Plan deadline and the 60-day OMB memorandum revisions. via Federal Register
  • OpenAI updates Stargate with RFP for land/power and RFQ for A/E firms. The $500B four-year buildout moved into procurement: a Request for Proposals for land and power partners and a Request for Qualification for architecture/engineering firms providing site design services. via OpenAI

Who shipped

OpenAI‘s o3-mini was the headline drop. Anthropic, Google DeepMind, Meta, and xAI stayed quiet on the product front. The o3-mini System Card was published alongside the model release, including Safety Advisory Group classification of the pre-mitigation model as Medium risk under the Preparedness Framework. via OpenAI

Open-source pulse

DeepSeek hit roughly 22 million DAU and around 125 million cumulative users — more than 100 million of those added in the final week of January. Hugging Face’s Open-R1 reproduction effort crossed 10,000 GitHub stars within three days of its Jan 28 launch (768-H100 Science Cluster). Alibaba’s Qwen2.5-VL family continued landing in benchmark replications against GPT-4o and Claude 3.5 Sonnet. via Hugging Face

Money, infra & hardware

NVDA closed January below pre-DeepSeek levels after the Jan 27 single-session record loss. Trump reinforced threats of up to 100% tariffs on Taiwan-made semiconductors; Taiwan signaled an industrial support package. Microsoft maintained guidance of approximately $80 billion in FY25 AI-datacenter capex, pushing back on the post-DeepSeek narrative that hyperscaler AI spending might be rationalised; the Big Four’s combined 2025 capex was tracking toward a record. via Washington Post

By the numbers

  • $1.10 / $4.40 — o3-mini API per-million-token pricing
  • 87.3% — o3-mini AIME 2024 pass@1 with high reasoning effort
  • 49.3% — o3-mini SWE-bench Verified (OpenAI’s then-highest)
  • ~22M / ~125M — DeepSeek DAU / cumulative users
  • 10K — Open-R1 GitHub stars in three days
  • Most-mentioned lab: OpenAI

Compiled by AI Feed’s editor from verified web sources for 31 January 2025.