OpenAI just published a playbook for trustworthy third-party evaluations of frontier models, warning that simple prompt-response testing is no longer sufficient. The company emphasized that surrounding scaffolds and tooling drastically shift performance, urging evaluators to specify claims and rigorously check for reward hacking, sandbagging, and test contamination.
OpenAI just explained why its models started obsessing over goblin and gremlin metaphors starting with GPT-5.1. A post-mortem traced the verbal tic to reward training for ChatGPT's "Nerdy" personality, which favored creature words and accidentally leaked into the broader model via RL feedback loops. OpenAI retired the persona in March and cleaned its training data.
OpenAI just expanded its Trusted Access for Cyber program and introduced GPT-5.4-Cyber, a fine-tuned model built specifically for defensive cybersecurity work. The permissive variant lowers refusal boundaries on legitimate security tasks and adds binary reverse engineering to analyze compiled binaries for vulnerabilities. It allows verified defenders to leverage frontier capabilities without running into standard safety guardrails.
Cloudflare just integrated OpenAI’s GPT-5.4 and Codex harness directly into its Agent Cloud platform and Sandboxes. The move lets millions of enterprise customers build and deploy production-ready AI agents at global scale for tasks like customer support, system updates, and automated reporting. Why it matters: it pushes frontier models right to the network edge, making autonomous workflows faster and far more secure.
GitHub just launched Rubber Duck in experimental mode for Copilot CLI, a cross-model review agent that taps another model family to critique coding plans and implementations. When Claude orchestrates, GPT-5.4 steps in at key checkpoints to spot blind spots. On SWE-Bench Pro, pairing Sonnet 4.6 with Rubber Duck closed 74.7% of the gap to Opus alone.
OpenAI just closed a massive $122 billion funding round at an $852 billion post-money valuation, backed by Amazon, Nvidia, and SoftBank. The company is generating $2 billion in monthly revenue with ChatGPT topping 900 million weekly active users. OpenAI plans to pour the capital into computing power and roll ChatGPT, Codex, and agent workflows into a single AI superapp.
OpenAI just launched GPT-5.4 mini and nano, lightweight models built for high-volume workloads and tool-heavy coding. The mini variant runs over 2x faster than GPT-5 mini while nearly matching the flagship GPT-5.4 on SWE-Bench Pro, at just $0.75 per million input tokens. Nano drops the entry cost to $0.20, perfect for background subagents.
OpenAI just launched ChatGPT for Excel in beta, letting users build, analyze, and update spreadsheet models using plain English. Powered by GPT-5.4, the add-in preserves native formulas, asks for confirmation before editing workbooks, and traces logic cell by cell. It also introduces live financial data integrations with Moody's and Dow Jones Factiva.
OpenAI just launched GPT-5.4 across ChatGPT, the API, and Codex, packing native computer-use capabilities and a massive 1M-token context window. It tops human performance on desktop navigation with a 75% OSWorld score and cuts tool-calling token bloat by 47%. Best of all, you can now redirect its thinking process mid-response when a workflow drifts off track.
