Claude Sonnet 5 in the Wings, a Post-Mythos Model, Sakana's Fugu, and GPT-5.6 Pro This Week

It's a busy week for model launches and leaks. Anthropic is prepping the fifth Claude Sonnet, with the model slug surfacing through a partner provider. WorldofAI's early testing reports a solid upgrade with impressive outputs; some examples come later in the video.

More striking is the report of a model trained that's more capable than Mythos — pushing past the already-banned, already-controversial model. The release path matters: ship through Project Last Sync, fold into a future public model, or hold internally to accelerate the next generation. Anthropic's choice will say a lot about how it now reads the regulatory environment.

The competitor side is loud too. OpenAI is set to launch GPT-5.6 Pro this week — Tuesday or Thursday looks likely. Alongside it ships BDI (GPT BDI 1), the first major voice-model refresh since GPT-4o, with an August 2025 knowledge cutoff. Some users already see it rolling out in the ChatGPT app.

Japan also entered the frontier conversation. Sakana, a new AI lab, unveiled Fugu Ultra Systems and claimed Claude Fable 5 / Mythos 5-level performance on multiple benchmarks. WorldofAI's own look at the results says the claims are optimistic and the gap is real — but the model is still genuinely capable. A non-US challenger to the frontier oligopoly is worth tracking on its own terms.

The video runs an actual head-to-head: Sakana Fugu Ultra vs Claude Opus 4.8 Ultra Code, on a 3D Crossy Road-style game generation test. The split was striking.

Opus 4.8 Ultra Code won on final design quality, functionality, and polish. But it took 79 minutes, burned around 940K tokens, cost $37.85, and got stuck twice along the way. Fugu Ultra had real rough edges — inverted controls, a wonky camera, missing SFX — but finished in 22 minutes on ~90K tokens for $7.32 (roughly 1/5 the cost), and its difficulty scaling was actually better. The summary: Opus wins quality; Fugu wins speed, cost, and efficiency. That's now a real trade-off, not a theoretical one.