OpenAI Product Pulse

• Pricing hits $10 per 1M tokens for both input and output, 1567% above the $0.60 median for chat models and cheaper than just 3% of peers.
• Context...

GPT-5.5 delivers 1M-token context at $5/M input and $30/M output, while GPT-5 mini offers 128k context at $0.25/M input and $2/M output.

• Cost gap:...

OpenAI is curbing ChatGPT's sycophantic yes-man habit by retiring GPT-4o and rolling out models that cut overly agreeable responses by more than...

Impressive internal model gains mean little when lab-built tools remain buggy and bland. Outsiders consistently ship more capable software, proving that real-world demonstrations and deployed products are the only credible measure of progress.

OpenAI and Anthropic diverge sharply on consumer subscriptions versus enterprise economics.

• Subscription gap: Anthropic’s mid-tier Opus 5.5...

The article title signals dramatic mathematician reactions to OpenAI's reported math results drop, labeling them 'pure insanity'. Without further details on capabilities or expert validation, assessing genuine advances remains limited.

The OpenAI Agents SDK provides three primitives that move multi-agent systems from experiments to reliable production use.

• Handoffs let a triage...

Incredible operates directly in your signed-in browser and apps on Mac or Windows by voice command, while ChatGPT focuses on answers, writing, and...

AI agents like ChatGPT Work and Dots target everyday annoyances such as price monitoring, inbox sorting and trial reminders.

• Start with low-risk...

Governance and consistency matter more than raw output speed when RevOps teams adopt ChatGPT for battlecards, playbooks, and ICP docs. Build prompt...

• OpenAI publishes token-based rates for Chat, Work, and Codex models under Enterprise agreements
• Pricing measures input, cached input, and output...

ChatGPT's core smarts are advancing faster than its reliability and workflow design, creating daily friction for serious users.

• Default modes in...

Dots delivered fast automation in demos—building a full website in minutes and handling legacy app migrations at 95% accuracy—yet default corporate...

A brief outage left some ChatGPT workspace admins unable to view or manage plugin availability and settings, directly threatening enterprise...

• 文本生成视频：直接用文字描述创建视频，无需上传现有素材
• 角色动画化：可让先前生成的图像中角色动起来
• 工作流影响：标志 ChatGPT 从文本助手转向一站式创意制作平台，改变创作者流程

GPT Image 2.5 shows persistent quality and reliability problems, with legitimate image generation interrupted far more often by difficult filters than...

Agents are shifting from chat tools to operators with inboxes and purchasing power.

• Dedicated email access: Grok Bot and Instinct now use their own...

Agent2Agent (A2A) supplies the missing interoperability layer that lets specialized agents from different vendors discover each other, delegate tasks,...

Microsoft's Decision-1 shows specialized decision models gaining traction as a critical layer for powering LLM judges and screening research hypotheses, delivering faster, higher-quality results than general LLMs on structured tasks.

Small quality-of-life updates are making ChatGPT and Claude noticeably easier to weave into daily routines without fanfare.

• Audio uploads in...