AI Briefing, February 25 · Issue 59
GPT-5.3-Codex has officially launched across OpenAI's Responses API and OpenRouter, delivering 3–4× higher token efficiency and topping multiple programming benchmarks—including Terminal Bench. Meanwhile, Anthropic has released C...
Editorial standards and source policy: Editorial standards, Team. Content links to primary sources; see Methodology.
## 🔍 Key Insights
**GPT-5.3-Codex** has officially launched across OpenAI’s **Responses API** and **OpenRouter**, delivering **3–4× higher token efficiency**, and achieving top performance on multiple programming benchmarks—including Terminal Bench. Meanwhile, **Anthropic** has introduced **Claude Code Remote Control**, alongside its new **RSP 3.0 Responsible Scaling Policy**, while **OpenAI** and **Perplexity/Comet** are advancing next-generation agent interaction via **WebSocket acceleration** and the **gpt-realtime-1.5** model, respectively.
## 🚀 Highlights
- **GPT-5.3-Codex Now Generally Available**: OpenAI has released the model to all developers via the Responses API, enabling high-precision coding and intelligent agent task orchestration.
- **GPT-5.3-Codex on OpenRouter**: As a cutting-edge agent programming model, it is now accessible to third-party platform developers through OpenRouter.
- **Claude Code Introduces “Remote Control”**: Enables seamless terminal session handoff across devices (desktop ↔ mobile); the `claude rc` command is now available in research preview for **Claude Max users**.
- **OpenAI Responses API Adds Native File Support**: Now natively parses **docx, pptx, csv,** and **xlsx** files—significantly improving document context retrieval accuracy.
- **Perplexity & Comet Launch Upgraded Hands-Free Voice Mode**: Powered by OpenAI’s new **gpt-realtime-1.5**, tool invocation stability improves by **25%**, with full-browser hands-free control support.
- **Notion Workers™️ Alpha Released**: Notion introduces programmable CPU infrastructure, providing a runtime foundation for custom AI agents within the Notion platform.
- **OmniDocBench Evaluation Framework Called for Reform**: Jerry Liu argues that current document-understanding benchmarks have become **saturated and overfit**, urging a shift toward semantic correctness–driven evaluation standards.
- **Anthropic Releases RSP 3.0 and Security Resource Center**: Updates its “Responsible Scaling Policy,” consolidating all safety commitments, historical reports, and compliance practices to strengthen AI safety transparency.
GPT-5.3-Codex has officially launched across OpenAI’s Responses API and OpenRouter, delivering 3–4× higher token efficiency, and achieving top performance on multiple programming benchmarks—including Terminal Bench. Meanwhile, Anthropic has introduced Claude Code Remote Control, alongside its new RSP 3.0 Responsible Scaling Policy, while OpenAI and Perplexity/Comet are advancing next-generation agent interaction via WebSocket acceleration and the gpt-realtime-1.5 model, respectively.
🚀 Highlights
- GPT-5.3-Codex Now Generally Available: OpenAI has released the model to all developers via the Responses API, enabling high-precision coding and intelligent agent task orchestration.
- GPT-5.3-Codex on OpenRouter: As a cutting-edge agent programming model, it is now accessible to third-party platform developers through OpenRouter.
- Claude Code Introduces “Remote Control”: Enables seamless terminal session handoff across devices (desktop ↔ mobile); the
claude rccommand is now available in research preview for Claude Max users. - OpenAI Responses API Adds Native File Support: Now natively parses docx, pptx, csv, and xlsx files—significantly improving document context retrieval accuracy.
- Perplexity & Comet Launch Upgraded Hands-Free Voice Mode: Powered by OpenAI’s new gpt-realtime-1.5, tool invocation stability improves by 25%, with full-browser hands-free control support.
- Notion Workers™️ Alpha Released: Notion introduces programmable CPU infrastructure, providing a runtime foundation for custom AI agents within the Notion platform.
- OmniDocBench Evaluation Framework Called for Reform: Jerry Liu argues that current document-understanding benchmarks have become saturated and overfit, urging a shift toward semantic correctness–driven evaluation standards.
- Anthropic Releases RSP 3.0 and Security Resource Center: Updates its “Responsible Scaling Policy,” consolidating all safety commitments, historical reports, and compliance practices to strengthen AI safety transparency.