AI Product Updates: New Features and Releases

GPT-6 now has higher cache-hit rates by default, so more input tokens get cached-input discounts of up to 90%, helping agents run faster and cost less.

Developers can track cache-hit rates in a new Prompt Caching Dashboard and use a diagnostics API to find changes that blocked cache reuse and estimate how many tokens were affected.

With GPT-6 and the Motion MCP, users can turn an idea, assets and a visual reference into animated scenes, typography and transitions, then describe changes to have the video updated.

Users can manually reset their accumulated usage allowance, which Anthropic frames as a way to "explore Opus 5.5" and which observers see as a test feature.

Users can authorize AI agents to research markets, pay for data, and execute stock and ETF trades directly.

The startup's AI agent operates locally across Windows, Mac, and Linux devices.

The aggregation platform brings in three new models at once, usable with Capy credits, Codex/SuperGrok subscriptions, or a bring-your-own-key setup.

OpenAI adds mid-tier Sol and lightweight Luna to the GPT-6 family, with API prices half the GPT-5.6 promotional rate and prompt caching that survives mid-run reasoning-effort changes, available in ChatGPT Work and Codex.

Ramp AI Index gets a major update that now shows token and spend volumes, not just shares or token counts.

Higgsfield's API lets developers add its video generation tools, including Genjutsu, Cinema Studio 4.0 and Soul, into their own apps.

The TTS evaluation platform adds Controlled Voice Arena leaderboards for 9 new languages, letting users compare leading text-to-speech models by language.

Artificial Analysis released text-to-speech arena leaderboards covering nine languages beyond English, including Japanese, Chinese, Hindi and Spanish, where users can vote and compare model performance.