AIが期間内の動向を整理
[Sep 30] OpenAI speeds up generation, Google improves voice conversation, and Anthropic updates safety measures
September 30 saw progress on three fronts in generative AI: speed, more natural conversation, and safer operation. OpenAI announced faster generation for long outputs, Google improved voice-and-video conversation experiences, and Anthropic published a report on Claude misuse defenses to help users better understand real-world risks.
SOURCE CHECK
Primary Sources 7
Primary Sources
Key Points
- 1OpenAI announced the Ultrafast speed tier, making long tasks quicker to process.
- 2Google released Gemini 3.8 Live with Live Avatar, strengthening voice-and-video conversation.
- 3Anthropic summarized misuse detection and disruption cases for Claude, providing reference material for safer operations.
- 4OpenAI also updated GPT-6 pricing, making workload-based cost estimates easier.
OpenAI: Faster generation for long-running work
OpenAI announced an Ultrafast speed tier for ChatGPT / Codex. It is described as up to 8x faster token generation in Codex and up to 6x in the API, which helps reduce waiting time for long outputs and complex processing. This is especially relevant for workflows that involve many prompt iterations.
Google: A more natural voice-and-video conversation experience
Google's Gemini 3.8 Live with Live Avatar combines low-latency video generation with voice conversation. It helps keep conversations flowing and makes it easier to build interfaces that combine speaking, seeing, and listening. This is well suited to enterprise conversational services that need natural responses and visual guidance.
Anthropic: Understanding misuse cases to improve safety operations
Anthropic published a report summarizing cases where attempts to misuse Claude were detected and disrupted. It also covers trends over the past eight months and changes in defenses, making it useful for understanding risk from the user's side. The report can help when reviewing usage policies, monitoring, and access design.
OpenAI also made pricing easier to evaluate
OpenAI updated GPT-6 pricing and clarified standard, long-context, and fast-processing rates as well as data-residency handling. Beyond feature improvements, the pricing structure is now easier to read. For larger projects or ongoing use, this makes it easier to compare performance and operating costs.