Wednesday, September 2, 2026

Several pricing adjustments were detected, with Claude Code's floor increasing from $17 to $20/month and moving to a hybrid model. GitHub Copilot's floor also increased from $10 to $15/month. Conversely, Cursor's floor pricing shifted from $20 to $0/month, and OpenAI Codex saw its typical pricing shift from $100 to $20/month, though its floor increased from $0 to $8/month. Kimi Code released version 0.40.0, adding a Plugins panel to Settings, enabling the subagent model pool by default, and making the kimi acp subcommand run on the default agent engine. Claude Fable 5.1 https://news.ycombinator.com/item?id=49525473 and Claude Mythos 5.1 https://news.ycombinator.com/item?id=49525378 were also noted.

Gemini introduced agentic video understanding across its latest models, including Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite, to improve efficiency and reduce token costs when processing long videos https://www.youtube.com/watch?v=ytjgy30Cono. Google DeepMind's SVP Koray Kavukcuoglu discussed these frontier models, coding agents, and building AGI. GitHub Copilot announced its SDK for building custom agent capabilities and extending AI workflows across various programming languages, alongside the ability to build custom workflows with canvases in the GitHub Copilot app using a /create-canvas skill. OpenAI highlighted how AI-native companies utilize AI agents to enhance workflows, and noted that ChatGPT can now connect EHR and industry data for healthcare organizations.

In community developments, "Supafork" was introduced as a platform for sharing and forking AI agent sessions, supporting agents like Claude Code, Codex, and Gemini CLI. "Orthogonal" launched as an integration for agents to discover and pay for APIs, starting with over 50 API providers. The concept of AgentOps for automating SDLC, security, and pipelines was also discussed.

User experiences included a report of successfully using Fable to rewrite 65kLoC of Go in Rust for $400. Conversely, a user raised concerns that Claude Opus has declined in performance for complex tasks, citing issues with ignoring project rules and unverified edits, leading them to use Fable. A new method called "Semantic Overlays" was introduced to mitigate prompt injection for LLMs, demonstrating improved scores for Qwen-3.5-9B.