Jason Says
The biggest story today is Google I/O — Gemini 3.5 Flash went GA with no rate limits, and developers say it outperforms Sonnet at a lower price. Google is no longer 'announce and disappoint.' The AI landscape is genuinely reshuffling.
Google I/O 2026 unveiled the Gemini 3.5 model family, emphasizing frontier intelligence and agentic capability. Gemini 3.5 Flash is now generally available with no rate limits, priced roughly 3x its predecessor but still well below GPT-5.5 and Claude Opus 4.7. Early developer reviews praise strong instruction-following, standout agent coding performance, and a model that isn't simply benchmark-optimized.
Google I/O announced conversational AI inboxes for Gmail, voice capabilities in Docs and Keep, and a background 'information agent' in Search that proactively pushes topic updates. AI Mode, one year in, has measurably shifted user behavior from keyword queries toward natural-language sentences. Google is systematically pushing its entire product line toward an agent-driven paradigm.
Cursor launched Composer 2.5 on Product Hunt, upgrading its position as the most developer-favored AI coding tool. The release lands amid Google I/O and a wave of major model announcements, with expected enhancements targeting agent mode and multi-model switching. Existing Cursor users should take note of the improved workflow capabilities.
A developer frustrated by high Claude Code and Codex bills built this local CLI tool to scan session logs and pinpoint context bloat: oversized CLAUDE.md files, repeated tool outputs, broad repo exploration, and stale session state. For heavy Claude Code users, it offers a direct path to reducing API costs without changing core workflows.
This open-source Claude Code Skills plugin installs a complete academic research workflow — topic selection through drafting, review, revision, and finalization — in a single command. Compatible with Claude Code CLI, VS Code, and JetBrains on v3.7.0+, it stands out as one of the few vertically complete solutions in the growing Skills ecosystem for researchers and academic writers.
OpenAI announced efforts to integrate Content Credentials, SynthID, and its own verification tools into a unified provenance standard, helping users identify and verify AI-generated media. The initiative represents critical infrastructure for combating deepfakes and serves as a necessary compliance foundation for content platforms navigating AI-generated content at scale.
Running long-context LLMs is expensive mainly due to KV Cache memory, not compute. OSCAR proposes an offline attention-aware covariance estimation method that compresses KV Cache to 2-bit while preserving accuracy — effectively quadrupling the context length fitting in the same memory. A significant win for local deployment and cost-sensitive API users.
Supertone 开源 Supertonic TTS 引擎，基于 ONNX Runtime 实现超低延迟本地多语言语音合成，支持桌面、浏览器、移动端和边缘设备，完全离线无隐私顾虑。对需要在 Agent 或应用中集成语音能力但不想依赖云端 API 的开发者是实用选项。
Subscribe for daily AI updates + free playbook
📘 Subscribe Free