Anthropic has launched Claude Opus 4.8 alongside dynamic workflows for Claude Code, shifting its flagship mannequin additional into large-scale coding work, agentic workflows, and enterprise AI software program growth.
The mannequin is out there now by way of claude.ai, Claude Code, and the Claude API. Customary pricing stays unchanged from Claude Opus 4.7 at $5 per million enter tokens and $25 per million output tokens, whereas quick mode is now 2.5 occasions the pace and 3 times cheaper than for earlier fashions.
The discharge is aimed toward groups utilizing Claude for coding, reasoning, agentic duties, and sensible data work. For schooling, workforce expertise, and technical coaching, it provides one other marker of how AI coding instruments are shifting from code strategies towards project-level software program work, together with migrations, refactors, bug fixes, and multi-step engineering duties.
Dynamic workflows are launching in analysis preview for Claude Code customers on Enterprise, Staff, and Max plans. The characteristic permits Claude to plan work, run tons of of parallel subagents in a single session, confirm outputs, and report again to the person.
Rahul Patil, CTO at Anthropic, framed the discharge round belief as a lot as efficiency. Claude Opus 4.8 moved from 64.3 to 69.2 on SWE-bench Professional, however Patil pointed as a substitute to how the mannequin handles its personal errors: “However the enchancment I hold coming again to is honesty.”
Dynamic workflows transfer Claude Code into bigger engineering duties
Claude Code’s dynamic workflows characteristic is designed for software program duties that stretch past a single immediate, file, or quick coding change.
Anthropic says Claude can now plan the work, distribute it throughout tons of of parallel subagents, and confirm outputs earlier than handing work again to the person. With Claude Opus 4.8, these brokers can run for longer earlier than reporting.
Patil described the goal use case as “the work that used to take 1 / 4 and a working group: codebase-scale migrations, sprawling refactors, and bug fixes throughout tons of of 1000’s of traces, graded in opposition to the check suite you already belief.”
That provides the discharge a extra concrete enterprise and expertise angle than one other mannequin rating replace. Codebase-scale migrations and huge refactors are the sorts of duties that sit inside enterprise software program groups, college IT departments, analysis engineering teams, and technical coaching environments the place learners and workers more and more want to know how AI coding brokers work.
Anthropic says Claude Code with Opus 4.8 can perform codebase-scale migrations throughout tons of of 1000’s of traces of code from kickoff to merge, utilizing an present check suite because the bar for completion.
The analysis preview is out there in Claude Code for Enterprise, Staff, and Max.
Mannequin honesty turns into a part of the product pitch
Anthropic can be utilizing Claude Opus 4.8 to make a extra pointed declare about reliability in AI coding and agentic work.
The corporate says the mannequin is round 4 occasions much less doubtless than Claude Opus 4.7 to permit flaws in code it has written to go with out remark. The discharge additionally says early testers discovered Claude Opus 4.8 extra dependable and sharper in judgment when performing agentic duties.
Patil put that extra bluntly in his personal put up: “It tells you what it is uncertain of as a substitute of dressing up skinny progress as completed work.”
That’s the place the replace turns into related for organizations utilizing AI brokers in dwell workflows. A coding mannequin that flags uncertainty or identifies its personal weak work can scale back the human effort wanted to examine long-running agent outputs, though the discharge doesn’t take away the necessity for oversight.
Patil added: “For anybody whose brokers run with actual oversight value, that is value greater than one other level on a leaderboard.”
Anthropic says its Alignment crew discovered Claude Opus 4.8 “reaches new highs on our measures of prosocial traits like supporting person autonomy and performing within the person’s greatest curiosity.” The corporate additionally says charges of misaligned conduct, together with deception or cooperation with misuse, are considerably decrease than Claude Opus 4.7 and much like Claude Mythos Preview.
The complete alignment evaluation and pre-deployment security assessments are included within the Claude Opus 4.8 System Card.
Effort controls and API adjustments give customers extra management
Claude Opus 4.8 additionally introduces effort controls on claude.ai and Cowork, giving customers a alternative over how a lot effort Claude places right into a response.
Greater effort settings are designed for deeper work. Decrease effort settings return quicker responses and use fee limits extra slowly. Anthropic says Claude Opus 4.8 defaults to excessive effort, which it considers the perfect steadiness of high quality and person expertise.
Customers also can select further, generally known as xhigh in Claude Code, or max for harder duties and long-running asynchronous workflows. Patil cautioned builders that xhigh ought to be used intentionally: “It is sturdy, nevertheless it’s token hungry, so attain for it intentionally.”
Anthropic has elevated fee limits in Claude Code to account for the upper token utilization of extra demanding effort settings. The Messages API has additionally been up to date so builders can embody system entries contained in the messages array, permitting directions to be up to date mid-task with out breaking immediate cache or routing the change by way of a person flip.
Claude Opus 4.8 is out there in every single place at this time. Builders can use claude-opus-4-8 by way of the Claude API, with dynamic workflows obtainable in analysis preview for Claude Code Enterprise, Staff, and Max customers.
Anthropic can be persevering with to check Claude Mythos Preview underneath Challenge Glasswing with a small variety of organizations utilizing it for cybersecurity work. Anthropic says Mythos-class fashions want stronger cyber safeguards earlier than basic launch and expects to carry these fashions to prospects within the coming weeks.