BADA$$ AI: CREATE 100 WEBSITES/DAY
Claude Opus 4.8 delivers 69.2% SWE-Bench Pro, 4x fewer errors, and 3x cheaper fast mode. Explore the breakthrough AI model built for enterprise and developers.
Anthropic has once again raised the bar in artificial intelligence with the release of Claude Opus 4.8 on May 28, 2026. This isn't just a minor iteration; it represents a philosophical shift in how AI models should operate, focusing as much on honesty and reliability as on raw processing power. Available now on claude.ai, the Anthropic API, and major cloud platforms like Amazon Bedrock and Google Vertex AI, Opus 4.8 is designed to be the ultimate collaborator for complex coding, agentic workflows, and professional knowledge work.
Claude Opus 4.8 arrives just 41 days after Opus 4.7, showcasing Anthropic's accelerated release cadence in the face of fierce competition from models like GPT-5.5 and Gemini 3.1 Pro. The benchmark results are impressive:
Coding & Agentic Tasks: It achieves 69.2% on SWE-Bench Pro, surpassing GPT-5.5 and Gemini 3.1 Pro, and scores 88.6% on SWE-bench Verified.
Multi-Step Reasoning: The model excels in agentic reasoning, scoring 74.6% on Terminal-Bench 2.1 and an industry-leading 83.4% on OSWorld-Verified for computer-use tasks.
Knowledge Work: It set a new high of 1890 Elo on GDPval-AA, demonstrating superior performance in professional knowledge work and long-horizon tasks.
The core differentiator is its performance on long-horizon tasks. Independent analysis shows that while models like GPT-5.5 are faster on quick edits (tying near 90% on tasks under 15 minutes), Claude Opus 4.8 pulls significantly ahead on tasks lasting 1-4 hours, achieving a 74% success rate compared to 50% for GPT-5.5.
Perhaps the most groundbreaking aspect of Claude Opus 4.8 is its dramatic improvement in "honesty" and calibration. Early testers reported that the model is significantly more likely to flag uncertainties and less likely to make unsupported claims. This is backed by concrete internal evaluations:
4x Less Likely to Ignore Errors: The model is approximately four times less likely than Opus 4.7 to let flaws in its own code pass without flagging them.
Near-Zero Overconfidence: It achieves a 10x+ reduction in overconfidence, often opting to abstain from answering rather than guessing incorrectly.
Prosocial Alignment: The model reached new highs in prosocial traits like supporting user autonomy, with rates of misaligned behavior (like deception) substantially lower than its predecessor and similar to Anthropic's best-aligned model, Claude Mythos Preview.
Users now have a slider on claude ai to control how much effort Claude puts into a response. Higher effort settings mean more thinking time for complex tasks, while lower effort settings result in faster responses and slower rate-limit consumption. Opus 4.8 defaults to "High" effort.
Available in Claude Code for Enterprise, Team, and Max plans, this feature allows Claude to spawn hundreds of parallel subagents in a single session. The orchestrator plans the work, the subagents execute, and the results are verified before reporting back. This enables codebase-scale migrations across hundreds of thousands of lines of code.
Anthropic has significantly reduced the cost of fast mode. It runs at 2.5x the speed for $10 per million input tokens and $50 per million output tokens—three times cheaper than fast mode for previous models, making interactive, latency-sensitive applications more practical than ever.
The Messages API now accepts system entries inside the messages array, allowing developers to update Claude's instructions mid-task without breaking the prompt cache. This is a game-changer for long, agentic runs.
Independent testing by enterprise partners, such as Box, confirmed significant improvements in high-stakes tasks:
Report Drafting: 87% vs. 77% on industrial goods reporting tasks.
Legal Review: Caught more relevant clauses and flagged more potential issues in NDA reviews.
Financial Analysis: Extracted more accurate metrics in corporate lending analysis by nearly 8 percentage points.
Perhaps most compelling is the model's performance on the WorkBench revisited benchmark. Two years ago, the best agent completed 43% of tasks and caused unintended harm on 26% of them. Claude Opus 4.8 now completes 89% of tasks and takes unintended harmful actions on only 2.5%. Crucially, capability and safety improved together rather than trading off.
Claude Opus 4.8 is available today across all of Anthropic's surfaces and major cloud providers:
Standard Pricing: $5 per million input tokens / $25 per million output tokens (unchanged from Opus 4.7).
Fast Mode Pricing: $10 per million input tokens / $50 per million output tokens.
Context Window: 1M input tokens / 128K output tokens.
The primary difference is the model's improved "honesty" and calibration, making it 4x less likely to ignore errors. It also features better benchmark scores, a cheaper fast mode, and new features like dynamic workflows and effort control.
It depends on the task. On independent benchmarks, they are close in overall coding ability. However, Claude Opus 4.8 outperforms GPT-5.5 significantly on long, complex tasks lasting over an hour, while GPT-5.5 is slightly faster and more cost-effective on short, well-defined tasks.
Dynamic workflows allow Claude Code to act as an orchestrator, spawning hundreds of parallel subagents in a single session to plan and execute large-scale tasks, such as codebase migrations across hundreds of thousands of lines of code.
The standard pricing remains the same as Opus 4.7 at $5 per million input tokens** and **$25 per million output tokens. A faster mode is available at $10 per million input tokens and $50 per million output tokens, which is 3x cheaper than the fast mode of previous models.
Claude Opus 4.8 is available on claude.ai, the Claude API (using the model ID claude-opus-4-8), Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.
The model supports a massive 1 million token input context window, with a maximum output of up to 128,000 tokens, allowing it to process and analyze large documents and codebases in a single session.
Fast Mode is an optional setting that runs the model at roughly 2.5x the normal speed. It is priced higher than standard mode but is now significantly more affordable, costing $10 per million input tokens and $50 per million output tokens.
While the model is available, it is a paid tier. However, users on free plans may have limited access with reduced rate limits. Users also have control over "effort settings," where lower effort speeds up responses and uses rate limits more slowly.