Claude Code 2.1.208 Shrinks Session Transcripts 79x; Copilot's Gemini Models Deprecate July 31

Claude Code 2.1.208 ships 79x transcript compression and 7x faster tool-call processing today; GitHub Copilot's Gemini 2.5 Pro and Gemini 3 Flash retire July 31 — migrate to 3.1 Pro or 3.5 Flash before the deadline.

Claude Code 2.1.208 (July 14). The biggest performance release in weeks ships two wins directly relevant to long agent sessions: session transcript size drops up to 79x in edit-heavy sessions — the root cause of context exhaustion during large refactors — and per-tool-call CPU overhead is up to 7x lower at high tool counts. Both changes are automatic; no config needed. Additional fixes: CLAUDE_CODE_MAX_OUTPUT_TOKENS was silently ignoring scientific notation (1e6 was treated as 1 — update if you use env-var token limits), and very large markdown tables (200+ rows) no longer stall rendering. Update with claude update. Why it matters: Autonomous refactoring sessions that previously hit context limits mid-task can now run substantially further — the 79x compression directly extends how long Claude Code can work without a context reset.

GitHub Copilot: Gemini 2.5 Pro and Gemini 3 Flash deprecate July 31. GitHub will remove both models from all Copilot experiences on July 31, 2026 — chat, inline edits, ask mode, agent mode, and code completions. The migration targets are Gemini 3.1 Pro and Gemini 3.5 Flash. If your team's VSCode settings, workflow configs, or Copilot API integrations explicitly name either deprecated model, update before the deadline. Teams that haven't pinned a model aren't affected — the default updates automatically. Why it matters: Any agentic workflow that pins a specific Gemini model name will silently break on August 1.

Sources: Claude Code changelog — Anthropic · Upcoming deprecation of Gemini 2.5 Pro and Gemini 3 Flash — GitHub Changelog