Fast mode for Opus 4.7 on AI Gateway cuts latency 2.5x at 6x cost

Vercel has launched fast mode for Claude Opus 4.7 on its AI Gateway, achieving 2.5x faster output token generation while preserving the model’s full reasoning and intelligence [Vercel Changelog]. The feature is accessible by setting speed: 'fast' in the provider configuration for anthropic/claude-opus-4.7. Developers can also enable it globally via the CLAUDE_CODE_ENABLE_OPUS_4_7_FAST_MODE environment variable or in ~/.claude/settings.json, bypassing org checks with CLAUDE_CODE_SKIP_FAST_MODE_ORG_CHECK.

Pricing reflects the performance leap: input tokens jump from $5 to $30 per 1M, and output from $25 to $150 per 1M—six times the standard Opus 4.7 rate. Standard multipliers, including prompt caching discounts, still apply on top. The feature remains in research preview, labeled experimental by Vercel.

This trade-off—sharply reduced latency at a steep cost premium—targets latency-sensitive applications where speed outweighs budget constraints. Use cases include real-time code generation, interactive agents, and high-throughput inference pipelines where delays degrade user experience. At 6x pricing, fast mode is impractical for bulk processing or cost-sensitive deployments, but offers a lever for teams optimizing for responsiveness over efficiency.

Vercel positions this as a developer-controlled knob: speed can now be prioritized explicitly, with full transparency in cost and performance impact [Vercel Changelog].