A writing, coding and agentic-work assistant with a strong emphasis on long-form reasoning, controllability and safety.
Claude is the product family. Sonnet 5.5 is the current default for Free and Pro; Opus 5.5, Fable 5.1 and Haiku 5.5 complete the current 5.5 lineup.
Anthropic evolved Claude from a general conversational model into a family of fast Haiku, balanced Sonnet and frontier Opus/Fable models.
Sonnet 5.5 runs about 30% faster and cheaper per task than Sonnet 5, with built-in cyber safeguards and reduced cache read pricing for agent sessions.
Product features and underlying models are documented separately. Roadmap items include only provider-announced work; rumors are excluded.
Selected milestones that materially changed the product.
Anthropic rolled out the 5.5 generation with improved speed, lower costs, adaptive thinking and advanced cyber safeguards.
Anthropic made its assistant available after an earlier closed alpha and safety testing period.
The Opus/Sonnet/Haiku family added vision; 3.5 Sonnet then raised coding and reasoning quality while introducing Artifacts.
Claude gained hybrid reasoning and Claude Code, followed by stronger long-running agent and coding behavior in the Claude 4 family.
The latest Sonnet focuses on autonomous plans, tool use, coding and professional work at scale.
The comparison uses provider-published capabilities and named evaluations, not a generic marketing score.
| Dimension | Previous | Current | Measured difference |
|---|---|---|---|
| Agent behavior | Strong coding and tool use | Plans and runs browser/terminal workflows more autonomously | Sonnet class moves closer to recent Opus-class agent ability |
| Reasoning control | Extended thinking with supported effort controls | Effort levels tuned for longer agentic work | More flexible speed-versus-depth selection |
| Safety behavior | Sonnet 4.6 baseline | Lower overall undesirable-behavior rate in Anthropic tests | Provider reports safer agentic use; not a guarantee for every prompt |
| Tokenization | Earlier tokenizer | Updated tokenizer | The same input may use roughly 1.0-1.35x tokens depending on content |
Sonnet 5's official post compares multiple task-specific evaluations. Because scores and effort settings differ, describing the whole model as '2x stronger' would be misleading.
Anthropic says Sonnet 5 approaches the larger Opus 4.8 on important agentic evaluations at lower price.
Check evidenceThe updated tokenizer can map identical content to more tokens; this is a cost consideration, not a capability gain.
Check evidenceClaude's established long-context capacity supports large documents and codebases; effective recall still depends on task design.
Check evidenceWhat happens behind the button, where it helps and what to verify.
How it works: Places generated documents, code or interactive content in a separate working canvas so the output can be iterated alongside chat.
Best use: Drafting, prototyping and reviewing deliverables
How it works: An agentic coding surface reads a repository, edits files, runs commands and tests, and reports the resulting changes.
Best use: Repository-scale implementation, debugging and review
Watch: Review diffs and command permissions before accepting agent changes.
How it works: Groups chats, instructions and reference files; retrieval can fetch relevant passages when the knowledge set grows beyond the active context.
Best use: Reusable team context, research libraries and consistent writing
How it works: Allocates additional inference time before the final answer; API controls can balance latency, cost and depth.
Best use: Complex analysis, coding and planning
Status reflects the provider's latest public wording. “No public date” means there is no verified launch schedule.
Anthropic announced additional cloud-provider availability for supported programs after native-platform launch.
Official announcementThe Sonnet 5 launch centers on longer tool workflows and higher effort levels; these product surfaces continue to receive updates.
Official announcementAnthropic has not published a release date or fixed specification for Sonnet 6.
Provider announcements, help documentation and system information.
Anthropic · Mar 4, 2024
Anthropic · Feb 24, 2025
Anthropic · Jun 30, 2026
Last editorial verification: 2026-10-28. Prices, access limits, preview status and model routing can change after this date.