No blog post, no announcement tweet — GPT-5.5 quietly appeared in the API changelog. Key upgrades: 64k context (up from 32k), a new JSON Strict output mode that guarantees schema compliance, and meaningfully faster response times on long prompts. GPT-5.5-turbo replaces GPT-5 in most ChatGPT Plus conversations starting this week. Early evals from Simon Willison and others show a notable jump in multi-step reasoning.
OpenAI has partnered with Astral (the team behind uv and ruff, now the de facto Python package and lint toolchain) to co-develop AI-native developer tooling. Details are sparse, but the partnership announcement hints at deep integration between OpenAI's code models and Python's tooling layer — think AI-aware dependency resolution and lint rules trained on GPT-generated code patterns. A preview SDK is expected at PyCon US next week.
ElevenLabs expands beyond voice cloning into a full AI creative suite: script-to-video (up to 3 min), AI soundscape generation, podcast production with multi-speaker voice cloning, and a new "brand voice" feature that locks your audio identity across projects. The new suite consolidates their existing tools (ElevenLabs Studio, SFX, Voice Library) under one subscription. Creator plan starts at $22/month. The script-to-video pipeline is legitimately impressive — it rivals Sora clips for short-form content.
Manus, the autonomous AI agent that went viral in March for completing multi-day research tasks, exits invite-only and opens to everyone. The public version ships with a task marketplace (pre-built agent workflows for research, outreach, and data collection), API access, and a self-hosted option for enterprises. The web scraping and form-filling capabilities remain the standout features — no other agent handles dynamic SPAs as reliably. Free tier: 10 tasks/month. Pro: $29/month for 200 tasks.
A Show HN that hit #3 yesterday: Espa embeds directly into DBeaver and DataGrip as a plugin and provides context-aware SQL autocomplete and refactoring suggestions. Unlike QueryPilot (which generates whole queries), Espa works on partial queries as you type. The author fine-tuned a Mistral 7B model on a corpus of 2M public SQL queries. Open-source with a $6/month pro tier for cloud inference. Worth watching — the HN comments thread is full of database engineers saying they'd pay for this.