Claude Opus 4.8 Released: Faster, Cheaper Fast Mode, Dynamic Workflows, and Honesty Improvements
Anthropic released Claude Opus 4.8 today, upgrading Opus 4.7 with benchmark improvements across coding, agentic skills, reasoning, and knowledge work. The new version is available at the same price as the previous model.
Key Features
- Effort control on claude.ai — users can now set how much effort Claude puts into a task.
- Dynamic workflows in Claude Code — lets the model tackle very large-scale problems.
- Fast mode runs at 2.5× speed and is now three times cheaper than fast mode for previous models.
Benchmark Highlights
According to the announcement, Claude Opus 4.8 is the only model to complete every case end-to-end on Anthropic's Super-Agent benchmark, beating prior Opus models and GPT-5.5 at parity on cost. On Online-Mind2Web (computer-use and browser agent tasks), it scored 84% — a meaningful jump over both Opus 4.7 and GPT-5.5. On CursorBench, it exceeds prior Opus models across every effort level.
On the Legal Agent Benchmark, Opus 4.8 is the first model to break 10% overall on the all-pass standard. Early testers also reported improvements in tool-calling efficiency (fewer steps for same intelligence), citation precision, and token efficiency on retrieval workflows.
Honesty Training
Opus 4.8 introduces explicit honesty improvements — the model is trained to avoid making unsupported claims and to flag issues with inputs/outputs proactively. This translated into higher-quality analysis and better signal-to-noise ratio in testers' evaluations.
Pricing
Opus 4.8 is available at the same price as Opus 4.7. Fast mode pricing is 3× cheaper than previous fast mode pricing. Multimodal token cost is 61% cheaper than Opus 4.7 for Genie (Databricks' AI agent).
📖 Read the full source: HN AI Agents
👀 See Also
Opus 4.6 Medium vs Low: Performance Differences and Pricing
Opus 4.6 medium costs approximately 50% more than the low version but addresses significant laziness issues found in the low-powered model. The medium version sits between low and high in performance benchmarks.
Meta's MCI Tool Captures Employee Interactions for AI Training
Meta is installing tracking software called Model Capability Initiative (MCI) on U.S. employee computers to capture mouse movements, keystrokes, clicks, and occasional screen snapshots for AI model training. The data aims to improve AI's ability to replicate human computer interactions like dropdown menu selection and keyboard shortcuts.
Claude doubles usage limits outside peak hours for two weeks
Anthropic is temporarily doubling Claude usage limits outside peak hours for all plans. Weekdays outside 5–11am PT/12–6pm GMT get 2x usage, with weekends getting 2x usage all day.
Claude Opus 4.6 and Sonnet 4.6 now offer 1M context at standard pricing
Claude Opus 4.6 and Sonnet 4.6 now include a full 1M context window at standard pricing with no long-context premium, plus expanded media limits to 600 images or PDF pages per request.