Claude Opus 4.8
Claude Opus 4.8 is Anthropic's upgraded Opus model for agentic coding, computer use, reasoning, professional knowledge work, and long-running Claude Code workflows.
Claude Opus 4.8 Capabilities
Anthropic positions Claude Opus 4.8 as a sharper, more reliable Opus release with visible gains in agent tasks, code work, computer use, and professional analysis.
Sharper Agentic Coding
Opus 4.8 raises the official SWE-Bench Pro score to 69.2% and is described by early testers as better at asking clarifying questions, catching mistakes, and resisting unsound plans.
More Reliable Tool Use
Anthropic says Opus 4.8 uses tools more cleanly, follows instructions more consistently, and is less likely than Opus 4.7 to let flaws in its own code pass unremarked.
Stronger Computer Use
The official benchmark table reports 83.4% on OSWorld-Verified, ahead of Opus 4.7, GPT-5.5, and Gemini 3.1 Pro under Anthropic's reported setup.
Effort Control
Users can choose how much effort Claude spends. High effort is the default, while extra and max effort can spend more tokens for harder tasks and long-running async workflows.
Dynamic Workflows
Claude Code can plan very large tasks, run hundreds of parallel subagents in one session, verify the output, and report back for codebase-scale migrations and similar work.
Better Honesty And Alignment
Anthropic reports stronger honesty behavior, lower misaligned behavior than Opus 4.7, and prosocial-trait highs in its alignment assessment for this release.
Claude Opus 4.8 Benchmark Results
Anthropic's official chart compares Opus 4.8 with Opus 4.7, GPT-5.5, and Gemini 3.1 Pro across agentic coding, terminal coding, reasoning, computer use, knowledge work, and financial analysis.
Official figures include 69.2% on SWE-Bench Pro, 74.6% on Terminal-Bench 2.1, 83.4% on OSWorld-Verified, 1890 on GDPval-AA, and 53.9% on Finance Agent v2.
Dynamic Workflows For Large Tasks
Opus 4.8 launches with Claude Code dynamic workflows, a research-preview feature for bigger sessions that can coordinate many subagents and verify work before returning results.
Anthropic says dynamic workflows are available in Claude Code for Enterprise, Team, and Max plans, and Opus 4.8 agents can run for longer.
Claude Opus 4.8 Pricing
Claude Opus 4.8 keeps regular usage pricing unchanged from Opus 4.7 and adds lower-cost fast mode pricing for higher-speed Opus use.
Prices are from Anthropic's May 28, 2026 Claude Opus 4.8 announcement. MyClaw plan pricing is separate from provider API token pricing.
Claude Opus 4.8 vs Opus 4.7 vs GPT 5.5 vs Gemini 3.1 Pro
The official benchmark picture is clearest when Opus 4.8 is read as an incremental but broad upgrade over Opus 4.7, with particularly strong agentic coding, computer use, and knowledge-work scores.
| Benchmark | Opus 4.8 | Opus 4.7 | GPT-5.5 | Gemini 3.1 Pro |
|---|---|---|---|---|
| Agentic Coding: SWE-Bench Pro | 69.2% | 64.3% | 58.6% | 54.2% |
| Agentic Terminal Coding: Terminal-Bench 2.1 | 74.6% | 66.1% | 78.2% | 70.3% |
| Multidisciplinary Reasoning: Humanity's Last Exam | 49.8% no tools / 57.9% with tools | 46.9% no tools / 54.7% with tools | 41.4% no tools / 52.2% with tools | 44.4% no tools / 51.4% with tools |
| Agentic Computer Use: OSWorld-Verified | 83.4% | 82.8% | 78.7% | 76.2% |
| Knowledge Work: GDPval-AA | 1890 | 1753 | 1769 | 1314 |
| Agentic Financial Analysis: Finance Agent v2 | 53.9% | 51.5% | 51.8% | 43.0% |
Terminal-Bench 2.1 notes, OSWorld-Verified methodology updates, and additional evaluations are documented in Anthropic's Claude Opus 4.8 System Card.
Use Claude Opus 4.8 With Long-Running Agent Skills
Pair Opus 4.8 with skills built for coding agents, GitHub issue automation, repo-scale context, terminal supervision, session history, and team updates.
Coding Agent Skill
Delegate complex coding tasks to Codex, Claude Code, Pi, or OpenCode agents running in background processes with full file access.
Open Coding SkillGitHub Issues Skill
Auto-fix GitHub issues using parallel AI sub-agents that implement changes, open PRs, and respond to code review feedback.
Open GitHub Issues SkillOracle Skill
Bundle prompts + repo files into one-shot AI requests via the oracle CLI — with browser automation, sessions, and smart file filtering.
Open Oracle SkillTmux Skill
Remote-control tmux sessions by sending keystrokes and scraping pane output — essential for managing Claude Code and parallel worker sessions.
Open Tmux SkillSession Logs Skill
Search and analyze your OpenClaw conversation history across session JSONL files using jq and ripgrep.
Open Session Logs SkillSlack Skill
Control Slack from your AI agent — send messages, react with emoji, manage pins, and fetch member info.
Open Slack SkillCompare Claude Opus 4.8 With Other Models
Use these pages to place Opus 4.8 in context against Claude Sonnet 5, GPT 5.6 Sol, and MiniMax M3 for agent work.
Claude Sonnet 5
Compare Claude Opus 4.8 vs Claude Sonnet 5 when you want a lower-cost Anthropic model for coding, agents, and production workflows.
View Claude Sonnet 5GPT 5.6 Sol
Read the GPT 5.6 Sol page when your shortlist includes Claude Opus 4.8 vs GPT 5.5-style flagship reasoning and agent models.
View GPT 5.6 SolMiniMax M3
Consider MiniMax M3 when you need long-context multimodal agent work and want a different price-performance profile from Opus.
View MiniMax M3Claude Opus 4.8 FAQ
Claude Opus 4.8
Run Claude Opus 4.8 In A Hosted MyClaw Agent
Use MyClaw to pair Opus-level model access with managed runtime, files, tools, browser workflows, and persistent agent sessions.
Run Claude Opus 4.8 in MyClaw