Anthropic tests its next model under Claude Opus 5.5

Anthropic appears to be quietly testing another model in the wild. Over the weekend, several users on X noticed Claude Opus 5.5 producing results well beyond its usual level, suggesting some of that traffic is being routed to a new checkpoint. Interestingly, this comes just three weeks after Anthropic launched Opus 5.5 with lower API costs.

X user can was among the first to spotlight the potential new model in testing. They shared a recording of the output they got from a single-sentence prompt that had the model use Blender's MCP. Here's the stunning result:

Kol Tregaskes noted that Opus 5.5 seems to be routing to a new checkpoint "again" and shared a quick check: ask Opus 5.5, "Do you know VoxelBench? Do not search." If it answers, you're probably on the new model. This mainly boils down to fresh training data, not a smarter model per se.

Still, it's not just a couple of users claiming that they've encountered the new model. Another one, @kimmonismus, said their requests seem to be hitting the new model, wondering whether it's a fresh Opus 5.5 checkpoint or something like Opus 5.6, and added that it "looks really, really good."

Chetaslua also shared an animation called "The life of a Human" with every frame drawn in JavaScript. According to the post, it took two short prompts, about two hours, and roughly 27 million tokens at max thinking effort. He also mentioned that it could be Fable 5.5 or a new Opus model, possibly Opus 5.6.

He then shared another "Claude Next" demo, claiming the coded visuals are good enough to rival image and video generators.

Someone else also followed up with a 3D scene the model reportedly built from scratch with Blender MCP after a single-sentence prompt, working for two hours without ready-made assets, and called it the most impressive output they've had from any model.

There's a slight chance this is Fable 5.5, or whatever label Anthropic settles on. But since it's routing to Opus 5.5, an Opus 5.6 test seems at least as likely. Either way, the early results are incredible. Anthropic has kept up a fast pace lately, shipping Sonnet 5.5 and Haiku 5.5 in the past couple of weeks, and building its own voice models.

Opus 5.5 already costs $4 and $20 per million input and output tokens. If Anthropic can bring this level of output at a similar price, it would be a game changer. OpenAI recently pushed GPT-6.1 Sol to near-Astra performance at one-fifth of the price, and Google has Gemini 4 Argon showing up in A/B tests, but OpenAI in particular will have to put its best foot forward to keep up.

John Hoffmann, Partnerships & Outreach Manager at Anthropic, has commented on these outputs recently: "The best is yet to come. Stay tuned, peeps." Yet the model name and timing remain open. If you use Opus 5.5, you might already be trying it.