Suspicious output differences suggest Anthropic is quietly testing Haiku 5.5
legit_api · x · 2026-10-04
X user @notjazii claims Anthropic is internally testing Haiku 5.5, amplified by @legitapi. After days on Fable 5.5, they noticed some Fable 5.1 xhigh sessions (with the reset prompt known) produced outputs clearly worse than Sonnet 5.5.
- Their only explanation: Anthropic is routing some prompts to Haiku 5.5 through Fable 5.1 for testing
- They expect a Haiku 5.5 release within a week or two
- Evidence shown via side-by-side outputs on identical settings
Unverified community speculation, not an official announcement.
More from Models
- OpenAI's Mark Chen on hack fallout: 'We're not going to shoot ourselves in the foot' — pstAsiatech · 2026-10-04
- OpenAI's $200 plan usage effectively halved as compute crunch deepens, analysts note — kimmonismus · 2026-10-04
- Qwen 27B on a single R9700: 3-bit rotation quant buys 569K-token cache and 12x faster agent turns — evp-cloud · 2026-10-04
- User ditches Grok subscription: NSFW writing loses to DeepSeek, images trail rivals — Otherwise-Wrap7406 · 2026-10-04
- GPT 6.1 slammed for endless checks: simple doc task takes 1.5 days vs 1-2 hours on rivals — Simple-Law5883 · 2026-10-04
- o1-pro launched at $150/$600 per million tokens; frontier models seen at $0.40-0.80 by 2027 — AiBreakfast · 2026-10-04