Claude Opus 5.5 leaks with slashed pricing
Anthropic is reportedly conducting early-access testing on an unannounced Claude Opus 5.5 checkpoint under the internal codename "Claude Wafer EAP." Leaked benchmarks indicate GPT-6 Astra-level capabilities alongside significantly reduced API rates of $4 per million input tokens and $20 per million output tokens.
Anthropic is preparing to shift the frontier model battlefield from pure benchmark supremacy to brutal cost-to-performance efficiency.
- –Price disruption: Leaked pricing of $4 input and $20 output per million tokens significantly undercuts traditional frontier flagship rates, making Opus-class capability viable for continuous, high-volume agent loops.
- –Frontier parity: Targeting performance on par with next-generation models like GPT-6 Astra indicates that post-training and architectural optimizations have closed the gap between capability scaling and inference overhead.
- –Strategic timing: Stealth testing under "Claude Wafer EAP" signals that Anthropic is accelerating release cycles to preempt competitive flagship updates and keep developer loyalty locked into the Claude ecosystem.
DISCOVERED
1h ago
2026-09-21
PUBLISHED
1h ago
2026-09-21
RELEVANCE
AUTHOR
WorldofAI