Claude Capybara is not a public Claude model you can select today. As of March 28, 2026, Anthropic's public docs still list Claude Opus 4.6, Claude Sonnet 4.6, and Claude Haiku 4.5 as the current public Claude lineup. The Capybara name appears only in reporting around an unreleased model Anthropic says it is testing with a small group of early-access customers. If you need Anthropic's strongest public model right now, the official route is still Claude Opus 4.6.
“Evidence note: this guide reflects Anthropic's public models overview, pricing page, release notes, and Fortune's March 26, 2026 report on the leak and Anthropic's statement about the unreleased model.
TL;DR
- Not public today.
Claude Capybaradoes not appear in Anthropic's public model overview, pricing page, or release notes as of March 28, 2026. - Partly confirmed, mostly not. Anthropic has acknowledged testing a stronger general-purpose model with early-access customers, but it has not published Capybara as a public product contract.
- Capybara and Mythos are still reported names, not shipped names. Fortune's leak reporting ties
CapybaraandClaude Mythosto the same unreleased model, but Anthropic has not yet published a final public launch page saying which name will survive. - The strongest public Claude today is still Opus 4.6. If you need the highest-end Anthropic model now, use Opus 4.6. Do not design purchasing, integration, or roadmap decisions around Capybara yet.
- Wait for real launch signals. A public model alias, pricing row, release note, and official launch page are the signs that would turn this from rumor into product surface.
What Claude Capybara actually is today
The cleanest answer is also the most useful one: Claude Capybara is a reported name for an unreleased Anthropic model or model tier, not a current product you can use. That distinction matters because most of the confusion comes from mixing together three different layers of the story.
The first layer is public and straightforward. Anthropic's own documentation still shows a public top-end Claude line built around Opus 4.6, with Sonnet 4.6 and Haiku 4.5 underneath it. Those are real launch surfaces. They have public API aliases, public pricing, release notes, and a stable place in Anthropic's current product documentation.
The second layer is also real, but narrower. Fortune reported on March 26, 2026 that Anthropic acknowledged testing a new model with early-access customers after draft launch materials were exposed through a data leak. That means there really is a stronger unreleased Anthropic model in testing. The existence of a stronger in-flight model is no longer pure rumor.
The third layer is where people overreach. The same Fortune report says the leaked draft materials used the names Claude Mythos and Capybara, and described Capybara as a higher-end tier above Opus. That is important reporting, but it is still not the same as Anthropic publishing a public launch contract. A leaked draft can reveal direction. It does not give developers or buyers the things they actually need: a product page, model alias, pricing, access rules, or release date.
So if your real question is, "Can I use Claude Capybara right now?" the answer is no. If your question is, "Is Anthropic working on something stronger than Opus 4.6?" the answer is yes, based on Anthropic's own statement to Fortune. Keeping those two answers separate is the whole job of this article.
What Anthropic has actually launched right now
Before talking about Mythos or Capybara, it helps to anchor yourself in Anthropic's actual public model surface on March 28, 2026.
| Public Claude model | Public API alias | Position in Anthropic docs | Base pricing |
|---|---|---|---|
| Claude Opus 4.6 | claude-opus-4-6 | Anthropic's most capable public model for complex reasoning and coding | $5 input / $25 output per MTok |
| Claude Sonnet 4.6 | claude-sonnet-4-6 | Anthropic's speed-and-intelligence middle tier | $3 input / $15 output per MTok |
| Claude Haiku 4.5 | claude-haiku-4-5 | Anthropic's fastest lower-cost public tier | $1 input / $5 output per MTok |
That table is pulled from Anthropic's current public docs, not from community summaries. It is also the most important grounding correction in this whole discussion. If a name is not present alongside these public model families in Anthropic's model overview or pricing page, it is not yet a real buying or integration choice.
The release timeline tells the same story. Anthropic's public release notes show Claude Opus 4.6 launching on February 5, 2026 and Claude Sonnet 4.6 launching on February 17, 2026. Those dates matter because they show how recent Anthropic's last public flagship cycle already is. Capybara is not filling a long-standing public gap in the product line. It is appearing as a possible next tier or next flagship generation beyond a still-fresh public lineup.
That is why the most practical present-tense guidance is simple. If you need Anthropic's strongest public model today for coding, long-context reasoning, or high-end agent work, use Opus 4.6. If you need a better price-speed balance, use Sonnet 4.6. If you need a lower-cost fast model, use Haiku 4.5. Capybara may matter later, but it does not replace any of those choices today because it is not a shipping contract yet.
If you want the fuller public access and pricing picture around Anthropic's current top-end model, see our Claude Opus 4.6 pricing and subscription guide.
What the leak reporting says about Capybara and Claude Mythos
Fortune's March 26 report is the most important secondary source here because it narrows the gap between rumor and fact without pretending to close it completely.
The confirmed part is that Anthropic told Fortune it is developing a general-purpose model with advances in reasoning, coding, and cybersecurity, and that the model is being tested with a small set of early-access customers. That is a meaningful confirmation. It tells you this is not just fan fiction from a forum thread. Anthropic itself says the model exists and is under controlled testing.
The reported, but not yet officially productized, part is the naming and tier structure. Fortune says the leaked draft materials referred to a model called Claude Mythos and also described Capybara as a new tier above Opus. The report says the leaked materials made Capybara sound more capable and more expensive than Opus, and treated Mythos and Capybara as the same underlying system. That is a strong clue about Anthropic's intended positioning, but it is still a clue rather than a released contract.
This is where readers need to resist two bad habits. The first is dismissing the whole thing as fake because it came through a leak. Anthropic has already confirmed enough that the underlying development is real. The second is overcorrecting and acting as if the entire leaked launch narrative is now official. It is not. The final product could keep one name, change both names, soften the positioning, narrow the rollout, or take longer than the draft suggested.
The cybersecurity angle is a good example. Fortune says the leaked draft framed the model as unusually strong on cybersecurity tasks and described a cautious release approach because of that risk. That is newsworthy, but it still does not answer the product questions a real user has to solve. Even if every safety concern in the draft survives intact, you still do not know the public API alias, price, context window, usage limits, launch date, or access path. Safety headlines tell you why Anthropic may move carefully. They do not tell you what you can do today.
Why Capybara is not a real buying or build decision yet
People often confuse plausible future product with usable current product. The easiest way to keep those apart is to ask a boring question: what would a real public launch need that Capybara still does not have?
It would need a public model page. Anthropic's public docs already give Opus 4.6, Sonnet 4.6, and Haiku 4.5 explicit places in the product stack, along with API aliases and platform identifiers. Capybara has none of that in public docs on March 28, 2026.
It would need a public pricing row. Anthropic's pricing page is where real models become real budget decisions. That page already lists current Claude families and older public releases. There is no Capybara row there.
It would need a release note or launch post. Anthropic's release notes are not perfect, but they are where public availability becomes legible. Capybara does not appear there either.
It would need an access path. Is it API-first? Enterprise-only at first? premium plan only? invite-only for months? limited to defense or security customers? Right now the only guidance comes from reported draft language, not from a published customer contract.
Once you view the topic through that lens, the practical answer becomes much calmer. Capybara may be the next major Anthropic launch. It may even become a permanent tier above Opus. But until Anthropic publishes the contractual pieces that make a model actionable, there is no reason to rewrite architecture plans, buying decisions, or evaluation workflows around it.
This is also why a lot of social-post debate misses the point. Whether Capybara would beat Opus 4.6 on a benchmark is interesting, but it is still the wrong first question for most readers. The first useful question is whether Capybara exists as a public product. Right now, the answer is still no.
What to use right now if you need Capybara-level capability
If your real goal is not gossip but capability, the present-tense routing is simpler than the rumor discussion makes it seem.
Use Claude Opus 4.6 if you want Anthropic's strongest public model for complex reasoning, coding, and long agent loops. Anthropic's own model overview positions it as the latest high-end public Claude for the hardest tasks, and the public pricing page gives you a real cost structure you can plan against.
Use Claude Sonnet 4.6 if you want a more balanced cost-speed profile and do not need the top public tier on every task. For many teams, this is still the better production default because the difference between "best public model" and "best public production default" is usually cost and latency, not bragging rights.
Use Claude Haiku 4.5 if your real priority is speed or lower cost. The existence of a rumored super-tier above Opus does not suddenly make Haiku irrelevant. In fact, it makes the separation between research headline and actual deployment choice even more important.
The larger lesson is that you should optimize for the strongest available model that fits your workload, not for the most dramatic rumored model that might arrive later. Capybara may eventually reset Anthropic's public lineup. On March 28, 2026, it has not done that yet.
What signals would make Capybara real
If you want a durable way to follow this story without getting jerked around by speculation, watch for these five signals:
- A model overview entry. If Anthropic adds Capybara or Mythos to its public models overview, the story changes immediately.
- A pricing row. A real model eventually becomes a budget line.
- A public API alias. Anthropic's real public models get names developers can actually call.
- A release note or launch page. That is when access timing, tiering, and rollout become public rather than inferred.
- A clear access contract. Whether the model starts as API-only, Enterprise-first, early-access security program, or broader release matters more than the rumor headline itself.
Until those appear, the safest stance is neither denial nor hype. It is provisional interest. Anthropic appears to be testing something stronger than Opus 4.6. That matters. But for normal builders, buyers, and teams, the story stays incomplete until Anthropic publishes the contract.
FAQ
Is Claude Capybara the same thing as Claude Mythos?
Probably, but not in a fully official way yet. Fortune's report says the leaked draft materials tied Capybara and Claude Mythos to the same underlying model or launch. Anthropic has confirmed the existence of a stronger unreleased model, but it has not yet published which public name it will use.
Can I access Claude Capybara right now?
Not through Anthropic's public docs or public model lineup on March 28, 2026. The model Anthropic confirmed is being tested with early-access customers, not offered as a general public launch.
Is Capybara definitely a tier above Opus?
That is what the leaked draft materials reportedly said, but Anthropic has not yet published a public product page confirming the final tier structure. Treat it as reported positioning, not as a shipped public taxonomy.
When will Claude Capybara launch?
Unknown. No public Anthropic release note, pricing page, or launch page currently gives a public date.
Should I wait for Capybara instead of using Opus 4.6 now?
Only if your project can afford indefinite uncertainty. For real present-tense work, use the public model that exists today. On the Anthropic side, that still means Opus 4.6 for highest-end public access.
Final verdict
Claude Capybara is best understood as a confirmed development signal with an unconfirmed public contract. Anthropic has acknowledged that a stronger model exists and is being tested. What Anthropic has not done yet is publish the normal launch pieces that make a model real for the rest of us.
So the responsible reading is straightforward. Capybara might become Anthropic's next flagship tier. It might eventually sit above Opus. It might materially change how people compare frontier models later in 2026. But on March 28, 2026, it is still the name of a future-looking story, not the name of a public model you can build or buy against. If you need the best public Claude now, keep using Opus 4.6 and wait for Anthropic to publish the rest.