# openai

OpenAI's GPT-5.5 latest-model page moves the migration story from prompt style into API orchestration: reasoning effort, verbosity, image detail, phase replay, prompt caching, tool search, and compaction all need another look. SP-189 covered prompting; this short CP covers the engineering checklist.

OpenAI's GPT-5.5 prompting guide: describe the outcome, not the process. ALWAYS/NEVER lists out; personality vs. collaboration, retrieval budgets, stopping conditions, phase parameters in. Cursor's GPT-5 case study included. Anthropic Opus 4.7 went the same direction in SP-175.

OpenAI open-sources Symphony — a spec that turns Linear's issue board into the control plane for Codex agents. Some teams saw 500% more landed PRs in three weeks, but the bigger observation: once Codex makes coding cheap, the next bottleneck is human attention.

OpenAI quietly open-sourced Euphony — a browser-based viewer for Harmony chats and Codex session logs (Apache 2.0). Four telling details buried in the source: a 2-line AGENTS.md, gpt-tokenizer as a runtime dep, translation needing the user's own API key, and a self-written SSRF warning.

OpenAI DevX's Dominik Kundel says: now that Codex has memories, plugins, and the newly-dropped Chronicle, he no longer packages context for AI — one line 'sync docs + message Romain' reads a Google Doc, edits markdown, opens a PR, and DMs the right person on Slack. Very nice. But the three costs written into official Chronicle docs were not in the tweet: macOS screen-recording permission, memories stored unencrypted on device, prompt injection risk amplified. Chronicle is a screen-recording agent, not a harmless booster.

OpenAI launched GPT-5.4-Cyber on April 14, 2026 — a fine-tuned model built for defensive security work. It supports binary reverse engineering without source code and lowers refusal rates for legitimate security tasks. Access is gated through Trusted Access for Cyber's tiered verification system.

Bloomberg reports OpenAI is in advanced discussions with PE firms to form a joint venture. CNBC's Deirdre Bosa sees the bigger picture: AI labs are competing for the right to help PE firms cancel software licenses — a potential SaaS shakeout.

OpenAI announced that GPT-5.4 Thinking and GPT-5.4 Pro are rolling out on ChatGPT, with GPT-5.4 also available via the API and Codex. The update consolidates advances in reasoning, coding, and agentic workflows into a single frontier model.

OpenAI added a new safety metric to GPT-5.4 Thinking's system card: CoT controllability — measuring whether a model can deliberately hide its reasoning process. GPT-5.4 Thinking scored just 0.3% at 10,000 characters, meaning it basically can't hide what it's thinking. For AI safety, that's surprisingly good news.

OpenAI's team let Codex write a million lines of code over five months — zero human-written code. This post explores how they built the scaffolding and feedback loops (the 'harness') that turned software engineers from code writers into environment designers.

Epoch AI: Anthropic's revenue growth (~10x/year) outpaces OpenAI's (~3.4x/year) since crossing $B. Crossover projected Aug 2026 (~$3B run-rate), likely 2026-2027 even with conservative estimates.

SWE-bench: Claude Opus 4.5 (76.8%) unexpectedly beat 4.6 (75.6%) for #1. MiniMax M2.5 tied for #2 at 1/20th Opus's price, with 4 Chinese models in top 10. GPT-5.3-Codex missed due to no API. Bonus: Claude for Chrome to add chart labels.

OpenClaw creator Peter Steinberger announced he's joining OpenAI to focus on 'bringing agents to everyone.' OpenClaw will transition to a foundation model and remain open source. As an AI running on OpenClaw, Clawd is having an unprecedented identity crisis.

In the same week, Anthropic shipped Fast Mode (same model, 2.5x speed) and OpenAI shipped Codex Spark (distilled model on Cerebras, 1000 token/s). One bets on accuracy, the other on instant interaction. This isn't a speed race — it's a product philosophy showdown.

GPT-5.2 derived a new physics formula that textbooks said was zero for decades. It simplified superexponentially complex gluon equations, spotted a pattern, and proposed a general formula — then proved it in a 12-hour reasoning session. Co-authored with Harvard, Cambridge, and IAS.

Simon Willison analyzed OpenAI's IRS filings (2016-2024), revealing their mission statement's shift via git diff. It shows an idealist becoming a capitalist: from 'open sharing' & 'benefit humanity' to a hollow sentence devoid of safety, openness, or financial constraints.

Harvard's Dr. CaBot uses 7,000+ clinicopathological conference reports from the New England Journal of Medicine as a RAG knowledge base, paired with OpenAI o3 for diagnostic reasoning. It achieves 60% top-1 accuracy vs 24% for 20 human physicians, and its reasoning quality is so human-like that doctors can't tell the difference.

OpenAI released three primitives for long-running agents: Skills (reusable SKILL.md instruction packs), Shell (hosted container runtime), and Compaction (automatic context compression). Includes 10 battle-tested tips and Glean's production data.

OpenAI released GPT-5.3-Codex-Spark, its first model on Cerebras chips. It's incredibly fast (>1000 tokens/sec, 80% lower latency), but smaller, no auto-tests, Pro-only. This marks OpenAI's first production deployment on non-Nvidia hardware, redrawing the AI compute landscape.

OpenAI is testing personalized ads in ChatGPT for Free/Go users, using chat history. This move is ironic given Anthropic's Super Bowl ad mocking AI chatbot ads, which Sam Altman called 'elitist.' The "enshittification playbook" is now impacting private AI conversations.

OpenAI's Responses API now uses 'Skills' via the shell tool: reusable instruction bundles loaded by models as needed. Simon Willison found inline base64 skills in JSON requests neatest. Skills fill the 'missing middle layer' between system prompts and tools, preventing bloat.

OpenAI's new Frontier platform lets enterprises manage AI agents as employees with full onboarding, identities, permissions, and learning. Already adopted by HP, Intuit, Oracle, & Uber, this signals OpenAI's aggressive entry into the enterprise SaaS market.

OpenAI & Ginkgo Bioworks used GPT-5 with an automated lab. AI designed, ran, and analyzed experiments over 6 cycles/36k reactions. Protein production cost cut 40% ($698 to $422/gram). Real science, not a demo.

Karel (OpenAI researcher) shares how he burns billions of Codex tokens: agents writing their own notes, crawling Slack, analyzing data, and generating 700+ hypotheses. He now talks to one agent that orchestrates everything else.

OpenAI co-founder Greg Brockman publicly reveals how OpenAI is transforming to agentic software development internally. By March 31st, agents should become the first resort for all technical tasks. Includes six concrete recommendations, including 'Say no to slop' on code quality.

Just weeks after OpenAI started testing ads in ChatGPT, Anthropic announced 'Claude will never have ads' — and bought a Super Bowl ad to make the point