Opus 5.5 matches Fable 5.1 performance while running 30% faster and costing 40% less per task than Opus 5. In Claude Code, 5-hour session limits increase 20% today, the model goes 25% further within those limits, and Pro, Max, and Team users receive a banked reset.
An Anthropic engineer ran a head-to-head: Opus 5.5 and Fable 5.1 both ported HAProxy from C to Rust, with both models passing nearly all tests. Opus 5.5 finished in 9.5 hours versus Fable 5.1's 12, and cost 51% less.
Anthropic's trq212 describes Opus 5.5 as a direct product of user feedback: clearer communication, lower per-token cost than Opus 5, Fable 5.1-level intelligence, and improved token efficiency across all effort levels. The launch also includes higher 5-hour rate limits and a banked usage reset.
Opus 5.5 is the first Opus model with Fable 5.1-level safeguards on cyber, bio, and frontier LLM development. Flagged requests automatically fall back to another model, and Anthropic says it is actively working to reduce false positives.
Opus 5.5 generated a 13,081-tile mosaic entirely in code with no image files, including animations where stars' reflections transform into 22 small fish. The demo highlights the model's generative coding capabilities.
An Anthropic team member used Opus 5.5 with a MAX subscription to run multiple redesigns of their personal website via workflows, having the model iterate and self-critique. It then assembled a video trailer showcasing all the design iterations.
A user discovered that Opus 5.5 can drive Blender to produce claymation videos from a single prompt on claude.ai, prompting a surprised reaction from an Anthropic team member.
Pro, Max, and Team users get a one-time limit reset available in Settings → Usage until Oct 22. Opus 5.5 is the new default for paid plans, and its lower price means both the 5-hour and weekly caps stretch 25% further.
Opus 5.5 ships with Fable 5.1 intelligence at a lower per-token price, expanded 5-hour rate limits, and a one-time banked reset for paid users.
The banked reset offered alongside the Opus 5.5 launch applies to both the 5-hour and the weekly usage limits.
A user argued that Opus 5.5's 40% per-token price cut is offset by higher token consumption per task, making it effectively more expensive. An Anthropic team member responded that the token usage chart needs to account for effort levels, noting that lower effort settings use far fewer tokens than MAX mode.
The near-verbatim prompt that generated a 13,081-tile animated mosaic: an 80-second square HTML film using WebGL2 and plain JavaScript, no libraries or asset files, with tiles that lift, flip, and fly like a flock.
Buried in the Opus 5.5 announcement blog post is confirmation that Sonnet 5.5 and Haiku 5.5 are also in development, and will carry over the same performance, efficiency, and safety improvements as Opus 5.5.
An Anthropic team member confirmed that the 'Claudish' style, the overly formal or distinctive phrasing users had complained about, has been removed from Claude's outputs.
Opus 5.5's distinctive writing style is a model-level default, not a setting. Users who want custom output styles can add them in .claude/output-styles.
Opus 5.5 is live in Claude Code and the Claude Platform, positioned as the new recommended daily driver for coding and agentic work given its intelligence, speed, and lower price.
Opus 5.5 drops to $4/$20 per million tokens on the Claude Platform, down from $5/$25 for Opus 5. Cache reads fall 60% to $0.20. Developers can run `/claude-api migrate` in Claude Code to switch to the new model.
Thinking is always enabled for Opus 5.5, with medium effort as the default and an adjustable `effort` parameter. For long-running agents, compaction summarizes older turns while preserving recent context. A fast mode is also available at up to 2.5x speed ($8/$40 per million tokens).
Opus 5.5 produces cleaner prose, uses less jargon, and more reliably follows user-specified writing rules. The improvement makes long Claude Code sessions noticeably easier to read and follow.
