- Replaces GPT-5.3 Instant across ChatGPT and ships in the API as `chat-latest`; old model stays available for paid users for three months.
- 52.5% fewer hallucinated claims on high-stakes medical, legal, and financial prompts; 37.3% fewer incorrect claims on user-flagged conversations.
- Responses are 30.2% shorter in words and 29.2% shorter in lines on internal evals.
- AIME 2025 jumps to 81.2 from 65.4; MMMU-Pro to 76 from 69.2.
- Personalization from past chats, files, and connected Gmail rolling out to Plus and Pro on web.
- Bloomberg: Extensions setting in iOS 27, iPadOS 27, and macOS 27 lets users route Siri queries to Claude or Gemini, not just ChatGPT.
- Ends OpenAI's exclusive Apple Intelligence deal that shipped with iOS 18.
- Selection lives under Apple Intelligence and Siri in Settings; requires the third-party app to be installed.
- Apple plans to unveil the change at WWDC on June 8.
- Government evaluation across nine benchmarks in cyber, software engineering, science, abstract reasoning, and math — including held-out PortBench and ARC-AGI-2 semi-private.
- V4 Pro scores 32% on CTF-Archive-Diamond cyber vs GPT-5.5's 71%; 46% on ARC-AGI-2 semi-private vs 79%.
- Performs roughly on par with GPT-5, which shipped about eight months earlier; called the most capable Chinese model evaluated to date.
- More cost-efficient than GPT-5.4 mini on 5 of 7 benchmarks, ranging 53% cheaper to 41% more expensive.
- Engineering deep-dive on rebuilding the realtime voice pipeline to keep first-hop latency, jitter, and packet loss low at scale.
- Replaces one-port-per-session media termination with a transceiver edge service that converts WebRTC into simpler internal protocols for inference, ASR, TTS, and tools.
- Smaller fixed UDP surface to load-balance and secure; stateful ICE and DTLS sessions get stable ownership across global routing.
- On HN front page; authored by Yi Zhang and William McDonald.
- Self-serve Ads Manager and partner buys roll out three months after ads went live in the US Free and Go tiers.
- Cost-per-click bidding plus expanded measurement; ads stay clearly labeled and don't influence answer ranking, OpenAI says.
- Plus, Pro, Business, Enterprise, and Education tiers remain ad-free.
01
OpenAI Ships GPT-5.5 Instant as ChatGPT's New Default Model
models openai.com
02
Apple Will Let iOS 27 Users Pick Their AI Provider for Siri
industry sherwood.news
03
NIST CAISI Pegs DeepSeek V4 Pro Eight Months Behind US Frontier
research nist.gov
04
OpenAI Details the WebRTC Stack Behind Voice AI for 900M Weekly Users
tools openai.com
05
OpenAI Opens ChatGPT Ads Manager Beta With CPC Bidding
industry openai.com