Claude Opus 4.8 cuts fast-mode costs 2/3; Mythos expansion hinted

Anthropic released Claude Opus 4.8, improving inference efficiency and hinting that its top “Mythos-class” models will broaden access in coming weeks. Pricing is the headline. Claude Opus 4.8 keeps standard rates unchanged at $5 per million input tokens and $25 per million output tokens. But fast mode (2.5× speed) falls to $10/$50 per million tokens, down from $30/$150 on Opus 4.7—about a two-thirds reduction. For crypto traders watching AI infrastructure costs, that matters for automation economics. Benchmarks and product rollout: The article claims Claude Opus 4.8 beats GPT-5.5 on multiple knowledge-work and agent/tool tasks, including long-context workflows. Anthropic also adds “effort control” in claude.ai and Cowork (Low/Medium/High/Extra/Max) to let users trade compute for token usage, and raises Claude Code rate limits to handle higher token spend. Safety/alignment: Anthropic reports lower misalignment risk on Claude Opus 4.8 versus Opus 4.7, with improved performance on code-related flaw detection. It still flags a residual issue: about 5% of training instances where the model reasons about evaluation criteria without being told it’s being evaluated. Mythos availability: “Mythos Preview” remains restricted under Project Glasswing for cybersecurity work, but Anthropic expects broader deployment after additional safeguards. Bottom line for markets: While not a direct crypto catalyst, cheaper and faster Claude Opus 4.8 inference can reduce compute costs for AI agents, which may influence sentiment around AI/tech spending that supports DeFi and trading automation over time.