Saturday, September 26, 2026

Several AI coding agents have seen recent pricing adjustments. Claude Code updated its typical pricing from $17 to $20/month and changed its model from hybrid to flat, with its floor price also shifting from $17 to $20/month. GLM 5.2 also transitioned its model from hybrid to flat, with its Coding Plan starting at 18 USD per month. Kiro adjusted its typical pricing from $40 to $20/month, with paid tiers ranging from $20 to $200 per user per month. In contrast, OpenAI Codex saw its typical pricing increase from $20 to $100/month, while its floor price shifted from $8 to $0/month. Individual plans for Codex include Free, Go ($8/month), Plus ($20/month), and Pro (from $100/month).

OpenAI Codex experienced outages, with multiple reports noting a "401 Outage" and "Issues with Codex – Identified – Full Outage." Users reported an "Incorrect API Key error" that appeared to be widespread, as mentioned in discussions like "Tell HN: Codex Is Down". The agent also saw a new 0.157.1 release, though release highlights were not determined. Community discussions highlighted new tools and functionalities, including Recurse, described as a "serverless agent harness" for developing and deploying specialist agents faster, using a "coding agent skill + a serverless execution runtime with a harness that takes in a system prompt + Python functions as tools."

Another tool, Ekselio, was presented as a local-first solution for finance workflows, orchestrating with an LLM and executing in the browser, and connecting to QuickBooks online and Fred. Jevmem offers "automatic project memory for Claude Code." Developers are also exploring wider integration, with Tui2web enabling terminal UI applications like Claude and OpenAI Codex to be mirrored to the web for mobile prompting, supporting Tailscale. A GitHub Copilot Day session highlighted VS Code's support for multiple agent harnesses and bring-your-own-key (BYOK) options, showing how a single GitHub Copilot subscription can work with Claude, Codex, and Copilot. An anecdotal report described a "coding agent pushed a commit deleting every file on main" in an incident report, while a case study from OpenAI reports that Proaction boosts sales by 60% and saves over 75 hours using Codex, GPT-Live-1, and GPT-6 Astra for fleet management. The "Efficiency-Throughput Gap with GitHub Copilot" was also a topic of discussion on Hacker News.