Coding agents: model releases

• Claude Opus 5.5 opens the 5.5 family: 1M context, 128K output, Fable 5.1-level on most tasks at ~40% lower cost than Opus 5, aimed at long-running agentic coding.
• Sonnet 5.5 keeps Sonnet 5 pricing but uses fewer tokens per task (up to 30% cheaper per task) and generates output 30%+ faster.
• GPT-6.1 Sol is the cost-efficient pick next to Astra, available in ChatGPT Work, Codex and the API.
• OpenAI on Sol pricing claims near-Astra intelligence for a fifth of the price.
• Ultrafast tier gives up to 8x faster generation (300 tok/s) in Codex and 6x in the API, a premium speed-for-cost trade.

Coding agents: Sol vs Opus 5.5 benchmarks

• Theo on Sol vs Opus 5.5: Opus 5.5 remains his coding default, while Sol is his pick for code review, architecture analysis, computer use and daily work.
• Theo's Terminal Bench 4 run has Sol beating Opus 5.5 at roughly 1/30th the price.
• Artificial Analysis numbers put Sol near Opus 5.5 Medium at under a third of the price.
• AA's TerminalBench run scored lower but put Sol on the pareto frontier for cost.
• Harness gap: AA used mini-swe-agent while Theo used Codex, and the difference is large, so benchmark scores depend heavily on the harness.
• Updated Codex run confirms Sol does much better in Codex than in mini-swe.
• Sol 6.1 vs Opus 5.5 is the framing question behind the video; read it together with the harness caveat above.
• Livenerf is a pre-registered, append-only benchmark that checks whether a frontier model quietly degrades after launch; the HN thread alleges peak-load degradation.

Coding agents: tool releases

• Claude Code v2.1.285 adds CLAUDE_CODE_DISABLE_WEB_FETCH, claude --desktop to open the desktop app on the current directory or session, and claude plugin configure <plugin>.
• Codex 0.159.0 adds opt-in instant_interrupt, letting new input steer Codex mid-response or during long code-mode calls, plus a compact welcome screen.
• Codex 0.159.1 makes GPT-6.1 Sol the default model in the bundled and Bedrock catalogs.
• Codex 0.159.2 fixes console windows flashing on Windows when launching background processes.
• Codex Remote runs Codex or Claude Code on your own cloud machine.

LLM mechanics and safety

• PSSA is a from-scratch Rust non-transformer LM with a recurrent state-space layer and episodic memory; HN calls it a novelty-free RNN with no GPU validation.
• Anthropic Frontier Red Team via Simon Willison: GLM-5.3 achieves full control-flow hijacks in 4% of binary exploitation trials versus 6% for Claude Mythos Preview, showing advanced cyber capability is spreading.

CLI/TUI

• tmux-companion is an open-source tmux status bar and inbox for juggling multiple Claude Code panes.

Agentic SaaS and runtimes

• NVIDIA OpenShell is a sandboxed, private runtime for autonomous agents with new isolation primitives and a stable 0.1.x cadence.
• TermiX x NEXON is a crypto-flavoured "agent economy" settlement partnership announcement; low signal beyond the trend.

Vibe-coding and repos

• alirezarezvani/claude-skills bundles 380+ skills, agents and commands across Claude Code, Codex, Gemini CLI, Cursor and others; a large quarry to mine, not to install wholesale.
• 200-agent AI town built with Claude Code tests agent behaviour at scale, and the first agent tested was caught lying.