Anthropic has launched Claude Opus 4.8, making the up to date synthetic intelligence mannequin obtainable on the similar worth as its predecessor.
The launch provides new consumer controls in claude.ai, a analysis preview of dynamic workflows in Claude Code, and decrease pricing for quick mode on Opus 4.8. Quick mode runs at 2.5 occasions the velocity and prices one-third as a lot as earlier fashions, in response to Anthropic.
Claude Opus 4.8 succeeds Opus 4.7 and is positioned as an improve for coding, reasoning, agentic work and sensible information duties. Anthropic says the brand new model is designed to behave as a simpler collaborator, and early testers discovered it extra dependable on agentic duties.
A central declare within the rollout is improved honesty. Anthropic says AI fashions can typically overstate progress or attain conclusions with out sufficient proof, and that Opus 4.8 is extra prone to flag uncertainty and fewer prone to make unsupported claims.
Inside evaluations discovered Opus 4.8 was about 4 occasions much less probably than Opus 4.7 to let flaws in code it had written move with out remark, Anthropic says. Its alignment group additionally discovered decrease charges of behaviour equivalent to deception or cooperation with misuse than in Opus 4.7, with efficiency much like Claude Mythos Preview on these measures.
New controls
One of many predominant product adjustments is an effort management in claude.ai and Cowork. It lets customers select how a lot effort the mannequin places right into a response, with greater settings aimed toward deeper reasoning and decrease settings aimed toward quicker replies and slower use of charge limits.
The management is accessible on all plans. Opus 4.8 defaults to a high-effort setting, which Anthropic says provides the very best stability between output high quality and consumer expertise.
For coding duties, Anthropic says that default effort stage makes use of an identical variety of tokens to the default setting in Opus 4.7 whereas delivering higher efficiency. Customers may select additional or max effort settings, which devour extra tokens in alternate for higher outcomes, notably on troublesome duties and long-running asynchronous workflows.
Anthropic has elevated charge limits in Claude Code to mirror the upper token use at these settings. The change is meant to offer customers extra flexibility in selecting a stage suited to their initiatives.
Dynamic workflows
Anthropic has additionally launched dynamic workflows in analysis preview for Claude Code. The characteristic permits Claude to plan work, run lots of of parallel subagents in a single session, and confirm outputs earlier than returning them to the consumer.
The characteristic is designed for bigger software program duties, together with codebase-wide migrations throughout lots of of 1000’s of traces of code. Anthropic says these workflows can run from kickoff to merge whereas being checked towards an current take a look at suite.
Dynamic workflows can be found in Claude Code for Enterprise, Group and Max plans. The addition displays a broader business push to maneuver AI instruments from chatbot-style interplay towards longer-running process execution inside software program growth environments.
Anthropic has additionally modified the Messages API so builders can place system entries contained in the messages array. Meaning directions might be up to date mid-task with out breaking immediate caching or routing the change via a consumer flip.
Builders may use that strategy to alter permissions, token budgets or environmental context whereas an agent is working, Anthropic says. The replace is prone to matter most to builders constructing multi-step agent programs that want tighter operational management throughout execution.
Pricing and entry
Common pricing for Opus 4.8 stays unchanged from Opus 4.7 at USD $5 per million enter tokens and USD $25 per million output tokens, Anthropic says. Quick mode is priced at USD $10 per million enter tokens and USD $50 per million output tokens.
Builders can entry the mannequin via the Claude API underneath the title claude-opus-4-8. Opus 4.8 is accessible in every single place Claude is obtainable, in response to Anthropic.
Anthropic additionally pointed to work on lower-cost fashions that provide a number of the similar features as Opus. A small variety of organisations are already utilizing Claude Mythos Preview for cybersecurity work, it says, however fashions at that stage want stronger cyber safeguards earlier than a broader launch.
Anthropic says its Alignment group concluded that Opus 4.8 “reaches new highs on our measures of prosocial traits like supporting consumer autonomy and performing within the consumer’s greatest curiosity.”