Frontier AI
claude-haiku-5-5 costs $0.10 input and $0.50 output per million tokens up to 100,000 tokens, and five times that above.
Why it matters. High volume work like classification, summaries and subagents gets far cheaper, but the new tokenizer counts about 30% more tokens for the same text.
Source: Introducing Claude Haiku 5.5
Haiku 5.5 returns a 400 error for budget_tokens thinking, assistant prefill, any top_k, and temperature or top_p away from their defaults.
Why it matters. Adaptive thinking is on by default, so select content blocks by type, or let /claude-api migrate in Claude Code make the changes.
Source: Claude Haiku 5.5 migration guide
Copilot CLI, the Copilot app and VS Code Agent Host sessions can limit the files, network and credentials that agent commands reach.
Why it matters. Admins can require sandboxing through enterprise managed settings that developers cannot weaken, at no extra cost.
Source: GitHub changelog
Cache reads on Sonnet 5.5 drop from $0.20 to $0.10 per million tokens, 0.05x the base input price, and cache writes are unchanged.
Why it matters. Agents and chat apps that resend long cached context get cheaper with no code change.
Source: Claude API pricing
The bundled and Amazon Bedrock catalogs default to GPT-6.1 Sol, and /mcp login signs in to an MCP server from a running session.
Why it matters. If you rely on the default model, your runs now use a different one, so pin a model where you need consistent results.
Source: Codex CLI 0.161.0 release
From CLI version 1.0.94-0, /model lists models from a running Ollama instance and adds one to the session without a restart.
Why it matters. Models must support tool calling and streaming, and choosing one does not turn on offline mode or stop GitHub telemetry.
Source: GitHub changelog