Claude Capybara is not a public Anthropic model you can use today. As of March 28, 2026, Anthropic's official documentation lists Claude Opus 4.6 as its most capable shipping model, complete with a public API identifier, published pricing, and documented limits. Capybara, by contrast, exists only in leak-backed reporting about an early-access system Anthropic is testing with a small group of customers. If you need to deploy a Claude model right now, the answer is straightforward: use Opus 4.6 .

Is Claude Capybara Actually Available to Use?

The short answer is no. On March 26, 2026, Fortune reported that Anthropic acknowledged developing a new model with early-access customers, describing it as a "step change" in capability and the most capable system the company has built to date. Some leak-backed draft material referred to the new system as Claude Mythos and described Capybara as a tier above Opus. However, this is not the same as a public launch contract .

There is no public Anthropic model page listing a Capybara API identifier, no public pricing card you can budget from, and no official launch announcement you can hand to procurement or engineering. This distinction matters because much model-comparison content quietly erases it. Once an unpublished system gets a memorable codename, people start discussing it as if it already has stable pricing, latency, rate limits, and a rollout path. None of that is public yet .

What Can You Actually Deploy Right Now?

Anthropic's public documentation is clear about what is available today. The current Claude models lineup includes:

• Claude Opus 4.6: Positioned as the most intelligent Claude model for complex work, with a 1 million token context window and 128,000 token maximum output
• Claude Sonnet 4.6: A mid-tier option for balanced performance and cost efficiency
• Claude Haiku 4.5: A lighter model for faster, simpler tasks

Opus 4.6 carries an API model identifier of claude-opus-4-6 and is priced at $5 per million input tokens and $25 per million output tokens. To put that in perspective, processing roughly 100,000 words of input would cost about $0.50, while generating 100,000 words of output would cost about $2.50. You can budget from this pricing, run evaluations against it, build internal guidelines around it, and route production traffic to it today knowing exactly which model you are calling .

Why Waiting for Capybara Could Freeze Your Project

Teams often make a subtle mistake when rumors about next-generation models circulate. They stop choosing a current model because a rumored future one sounds better, then spend weeks in planning limbo with no new deployment, no new evaluation baseline, and no new budget certainty. Unless your organization genuinely benefits from waiting on the next tier, that is usually a bad trade .

The trap is especially easy to fall into when a model has a memorable codename. Capybara sounds exciting, and the idea of a "step change" in capability is compelling. But the responsible reading is narrower: Anthropic appears to be testing something stronger than Opus 4.6, and Capybara is one name connected to that effort. That is real enough to monitor, but not real enough to plan against as if it were already in the model picker .

How to Make Smart Decisions About Claude Models Today

• Deploy Opus 4.6 if you need a model this week: You have a public pricing card for budgeting, a stable model identifier for evaluation, and documentation that engineering, finance, procurement, and security can all read without relying on media reports
• Keep model choice behind configuration: Do not bake claude-opus-4-6 into application logic or procurement documents more deeply than necessary. This lets you swap a future tier into the same evaluation harness instead of rewriting half the stack later
• Anchor your performance baseline to Opus 4.6: Test the model against your actual coding jobs, long-context workloads, or agent tasks. If Capybara launches and truly is a tier above Opus, your most useful question will be whether it beats your current Opus 4.6 baseline enough to justify the cost, not whether it beats a rumor
• Monitor Capybara only if you are scouting frontier models: If you maintain a frontier-model watchlist for your organization, run structured evaluations every time the top tier changes, or work in a domain where a genuine step-change in coding or cybersecurity performance would alter your model mix, then keep Capybara on your watchlist

The practical rule is simple: if your question is about what to deploy, buy, or test today, choose Opus 4.6. If your question is about what Anthropic might ask you to reconsider later, keep Capybara on the watchlist. But do not freeze useful work while waiting for details that are not public yet .

Starting with Opus 4.6 also gives you a meaningful internal benchmark for whatever comes next. When Anthropic eventually launches a Capybara-tier model publicly, you will want to compare it against your own Opus 4.6 workloads, not against whatever leaked scorecards happened to circulate on launch-rumor day. For most teams, the answer is straightforward: ship on Opus 4.6 now.