CLAUDE CODE V2.1.77 LIFTS OUTPUT CEILINGS TO 64K/128K AND FIXES LONG-SESSION PAIN; OFF‑PEAK USAGE TEMPORARILY DOUBLED
Anthropic raised Claude Code’s output token ceilings and shipped reliability fixes, while off‑peak usage limits are temporarily doubled. The latest Claude Code...
Anthropic raised Claude Code’s output token ceilings and shipped reliability fixes, while off‑peak usage limits are temporarily doubled.
The latest Claude Code release v2.1.77 increases the default maximum output for Opus 4.6 to 64k tokens and the upper bound for Opus 4.6 and Sonnet 4.6 to 128k. It also lands a long list of stability fixes that matter in real projects, including accurate cost tracking when the API falls back to non‑streaming, reduced memory growth in long sessions, CRLF line‑ending safety on Windows repos, and tighter permission handling. There’s a new allowRead sandbox option and a handy “/copy N” to grab older assistant messages.
Separately, Anthropic is temporarily doubling usage limits outside peak hours, which The New Stack covered here and Anthropic echoed on Instagram. If you batch heavy generations or evaluations, consider shifting them into those windows to stretch quotas.
Bigger outputs (up to 128k) enable full diffs, long docs, and multi‑file patches without awkward chunking.
Fixes to cost tracking, memory, and permissions reduce flaky runs and surprise bills in longer workflows.
-
terminal
Generate a long artifact (e.g., 40–80k token design doc or migration diff) and verify output completeness, latency, and cost behavior in streaming vs non‑streaming fallback.
-
terminal
Schedule batch codegen/review jobs off‑peak for a week and measure throughput per token budget during the doubled‑usage window.
Legacy codebase integration strategies...
- 01.
On Windows repos, re‑test code write/overwrite flows to confirm CRLF handling is fixed and diffs stay clean.
- 02.
Review enterprise permission policies: validate that deny/allow rules and the new allowRead behave as expected in sandboxed tasks.
Fresh architecture paradigms...
- 01.
Design workflows that assume multi‑chapter outputs (runbooks, schema docs, ETL plans) to exploit 64k/128k responses.
- 02.
Adopt skills/agent patterns from community guides to standardize prompts, verification, and reuse across services.