Leaked Claude Opus 5.5 skips to testing with $4/M input tokens, undercutting Opus 5 by ~20%
新智元 · wechat · 2026-09-21
Aggregated leaks (unconfirmed) suggest Anthropic skipped Opus 5.2 and is quietly testing Claude Opus 5.5, codenamed claude-wafer-eap, possibly launching this week:
- Capability: testers report a clear generational gap over Opus 5 in coding intuition, deep refactoring, and long-horizon planning, plus striking 3D generation demos; some claim it beats gpt-6-sol
- Pricing (rumored): $4/M input, $20/M output tokens (20% cheaper than Opus 5), with cache reads at $0.2/M — potentially cutting long-context agent costs by an order of magnitude
- Parallel track: Claude Fable 5.2 is also surfacing in early tests, praised for 3D scene generation, front-end aesthetics, and pure-code builds
- Context: comes as OpenAI's GPT-6 and Google's Gemini 4 Pro are also reportedly in gray testing, setting up a three-way clash this week
More from Models
- Codex PRO+ Users Report Usage Draining Much Faster Since Weekly Reset — Next_Technology6361 · 2026-09-21
- Open-source finetune project laya hits 6.9k stars with free Kaggle 2xT4 notebook — ojasvi_yadav · 2026-09-21
- jev-reranker edges out ruri-v3 on four of five Japanese retrieval benchmarks — amaarora · 2026-09-21
- New model's vuln stats look same as before: many found, few exploited in the wild — xeophon · 2026-09-21
- Tokenization isn't why LLMs miscount letters: models spell characters with 100% accuracy — maksym_andr · 2026-09-21
- Claude now makes typos in Russian, continuing output quality complaints — StewartalsopIII · 2026-09-21