Skip to content
Source · Daily Brief

AI Daily Brief — 14 April 2025

OpenAI’s biggest API-only release of Q2 lands the same day NVIDIA pledges $500B of US-built AI infrastructure. GPT-4.1 family puts coding SOTA, 1M context, and a nano tier into developers’ hands at sharply lower pricing. NVIDIA started Blackwell production at TSMC Arizona and supercomputer assembly at Foxconn Houston / Wistron Dallas — while the US government formalized H20 China export restrictions ‘for the indefinite future.’

Top stories

  • OpenAI launches GPT-4.1 family. Three API-only models — GPT-4.1, GPT-4.1 mini, GPT-4.1 nano — with major gains in coding, instruction-following and long context. All three support up to 1M tokens of context with a June 2024 knowledge cutoff. via OpenAI
  • GPT-4.1 sets new SWE-Bench Verified SOTA at 54.6%. A 21.4-point jump over GPT-4o (33.2%) and 16.6 points above GPT-4.5 (38%). Also hit ~52.9% on Aider’s polyglot diff benchmark — more than double GPT-4o. MMLU at 90.2%.
  • GPT-4.1 pricing undercuts GPT-4o. GPT-4.1: $2/M input, $8/M output. GPT-4.1 mini: $0.40/M input, $1.60/M output (~83% cheaper than GPT-4o while matching it on many evals). GPT-4.1 nano: $0.10/M input, $0.40/M output — OpenAI’s first nano model and its cheapest. Prompt-caching discount raised to 75%; Batch API adds another 50% off.
  • GPT-4.5 Preview slated for API deprecation on July 14, 2025. OpenAI notified developers GPT-4.5 Preview would be removed in three months, framing it as experimental and pointing users to GPT-4.1 as the migration path. Triggered developer backlash. via OpenAI Community
  • NVIDIA to manufacture AI supercomputers entirely in the US for the first time. Over 1M sq ft of US manufacturing space across Arizona and Texas. Blackwell chips started production at TSMC’s Phoenix fabs; packaging/testing with Amkor and SPIL in Arizona; supercomputer assembly with Foxconn in Houston and Wistron in Dallas. Mass production to ramp over 12-15 months. via NVIDIA
  • NVIDIA commits up to $500B of US-built AI infrastructure over four years. Jensen: ‘The engines of the world’s AI infrastructure are being built in the United States for the first time.’ Partners: TSMC, Foxconn, Wistron, Amkor and SPIL. White House later framed it as a ‘Trump Effect’ announcement tied to tariff pressure. via CNBC
  • USG tells NVIDIA H20 China export license is indefinite. The April 9 H20 export-license requirement is now ‘for the indefinite future’ — paving the way for the ~$5.5B charge NVIDIA disclosed the following day. H20 sales had been $4.6B in Q1 FY26 prior to the rule, with another $2.5B that could not be shipped. via NVIDIA 8-K
  • Windsurf gives away free unlimited GPT-4.1 for 7 days. Citing internal evals where GPT-4.1 scored 60% higher than GPT-4o on their coding benchmark, ~30% better at tool-calling and ~50% less likely to repeat edits. via Geeky Gadgets

Who shipped

OpenAI ran the model side. NVIDIA ran the manufacturing side. The US government set the policy frame with the H20 license formalization.

Open-source pulse

No major open-weights drop dated Apr 14. The GPT-4.1 release is API-only — OpenAI told consumers improvements would flow into ChatGPT’s GPT-4o over time. The split made GPT-4.1 a developer-targeted release rather than a consumer flagship, sharpening OpenAI’s enterprise positioning.

Money, infra & hardware

NVIDIA’s $500B / 4-year US-infrastructure pledge directly responds to the tariff pressure while locking in TSMC Arizona’s Blackwell production. The H20 license becoming indefinite represents a ~$5.5B realized loss for NVIDIA’s China business — the largest single AI-export-control hit of 2025 so far.

Quiet corners

No major open-source / China-lab releases dated Apr 14. No notable Anthropic, Google, Meta or xAI product news.

By the numbers

  • 54.6% / 90.2% / 52.9% — GPT-4.1 SWE-Bench Verified / MMLU / Aider polyglot
  • $2 / $8 — $0.40 / $1.60 — $0.10 / $0.40 — GPT-4.1 / mini / nano pricing per million tokens
  • 1M tokens — context window for all three
  • ~83% — GPT-4.1 mini price discount vs GPT-4o
  • $500B / 4 years — NVIDIA US-built AI infrastructure pledge
  • 1M+ sq ft — NVIDIA US manufacturing footprint
  • ~$5.5B — NVIDIA’s pending H20 China writedown
  • Most-mentioned company: OpenAI

Compiled by AI Feed’s editor from verified web sources for 14 April 2025.