Devs suspect Anthropic silently swapped Opus 5.5 for a new checkpoint after stunning code-render demos
新智元 · wechat · 2026-10-12
Developers widely suspect Anthropic is quietly routing some Opus 5.5 requests to an unreleased checkpoint (nicknamed "ClaudeNext," Opus 5.6, or Fable 5.5), flooding X with demos:
- Chetaslua generated a full life-story animation with 4 prompts, 2 hours, and 27M tokens — drawn frame-by-frame in pure JavaScript, no image/video models. He argues code-native visuals could make media generation models obsolete within six months.
- Side-by-side tests show dramatic leaps in 3D rendering, spatial reasoning, and material detail versus release-week Opus 5.5.
- A community "fingerprint test" circulates: ask Opus 5.5 at High reasoning "Do you know Voxel Bench? Do not search" — answering correctly suggests a newer checkpoint.
Context: Anthropic is sprinting toward a reported IPO, with investor valuations eyeing $2 trillion. The company has not confirmed any routing change.
More from Models
- Prepping for Claude Pro: home broadband, $10 VPS, used router — but virtual cards fail — lxfater · 2026-10-12
- AWS open-sources Strands Decider 2B, a tiny low-latency decision model — gleech · 2026-10-12
- Decision model leaderboard called out for omitting GPT-6 Sol and Luna Decisions — gleech · 2026-10-12
- All verifiable tasks are solved — only non-verifiable ones remain for AI — RaphaelDabadie · 2026-10-12
- Meta's DINOv3 ViT-L/16 feature extraction model trends on Hugging Face — facebook · 2026-10-12
- Checking 241 Pan-African Review articles with Pangram: AI flags only spike after 2025, countering detector bias claims — menhguin · 2026-10-12