Anthropic has expanded its Claude model lineup from three tiers to four, adding a new class called Capybara above the existing Opus flagship. The change was revealed through a data leak reported by Fortune on March 26, 2026, when internal draft materials described Capybara as “a new name for a new tier of model: larger and more intelligent than our Opus models.” This is the most significant structural change to Anthropic’s product lineup since the company introduced its three-tier system. The full tier system behind Claude Mythos reflects Anthropic’s long-term strategy.
What Is the Anthropic Capybara Tier
Capybara is not a model — it is a tier. The distinction matters. Just as Opus is a tier that contains specific models (Claude Opus 4, Opus 4.5, Opus 4.6), Capybara is a new tier that will contain its own models. The first and currently only model in the Capybara tier is Claude Mythos, which Anthropic describes as representing a “step change” in AI capability.
From Three Tiers to Four
Since its founding, Anthropic has organized Claude models into three tiers: Haiku (fast and cheap), Sonnet (balanced), and Opus (powerful and expensive). Each tier trades off differently between speed, intelligence, and cost. Adding Capybara above Opus creates a four-tier system: Haiku, Sonnet, Opus, Capybara — giving developers and enterprises a new option when they need the absolute highest performance available, regardless of cost.
The move mirrors a broader industry trend. OpenAI similarly expanded from standard GPT models to an “o-series” reasoning tier. Google introduced specialized Gemini configurations for different use cases. Anthropic’s approach differs because Capybara is explicitly positioned as larger and more expensive, rather than as a specialized variant. It aims to be the best at everything, not just specific tasks.
Why Anthropic Chose the Name Capybara
Anthropic’s tier naming follows a pattern rooted in music and nature. Haiku references the short Japanese poetic form — fitting for a fast, lightweight model. Sonnet references a more structured, medium-length poem. Opus means a major musical composition. Capybara breaks from the musical theme and leans into the animal naming convention already present in individual model code names.
Capybaras are the largest living rodents, native to South America and known for their calm, social nature. The choice communicates scale — this is the biggest, most resource-intensive class of model Anthropic builds. It also fits the company’s preference for names that feel approachable rather than intimidating.
Claude Mythos: The First Capybara Model
Claude Mythos launched the Capybara tier as its inaugural model. Leaked internal documents revealed Anthropic was deciding between “Mythos” and an alternative designation, ultimately settling on Mythos as the model name with Capybara as the tier. Anthropic confirmed its existence after the leak, calling Mythos “the most capable we’ve built to date” — a general-purpose model with breakthrough performance in coding, academic reasoning, and cybersecurity.
Anthropic’s Full Model Hierarchy Explained
Understanding where Capybara fits requires knowing what each existing tier offers and where its limits lie.
Haiku: The Lightweight Tier
Haiku is Anthropic’s fastest and most affordable tier. The current model, Claude Haiku 4.5, is optimized for high-throughput, low-latency tasks where speed matters more than depth. It handles text classification, simple summarization, and high-concurrency workloads efficiently. API access pricing sits at $0.80 per million input tokens and $4.00 per million output tokens — making it the entry point for cost-sensitive applications.
Haiku’s limitations become apparent with complex reasoning tasks. Multi-step analysis, nuanced code generation, and tasks requiring deep contextual understanding stretch beyond what this tier is designed to handle.
Sonnet: The Balanced Tier
Sonnet occupies the middle ground. The current Claude Sonnet 4.6 delivers strong code generation, content creation, and data analysis at a reasonable cost. Most developers building production applications default to Sonnet because it balances intelligence with affordability. Priced at $3.00 per million input tokens and $15.00 per million output tokens, it handles the vast majority of real-world use cases competently.
Sonnet struggles with the most complex architectural decisions, research-grade scientific reasoning, and tasks requiring extensive multi-step planning across large problem spaces.
Opus: The Former Flagship
Until March 2026, Opus was Anthropic’s best. Claude Opus 4.6 excels at complex architecture design, research report generation, difficult scientific reasoning, and tasks where getting the answer right matters more than getting it fast. At $5.00 per million input tokens and $25.00 per million output tokens, it serves as the premium option for enterprises and researchers who need the highest available quality.
Opus remains an excellent model. The introduction of Capybara does not diminish what Opus can do — it simply means there is now a tier above it for tasks that push beyond what even Opus can reliably achieve.
Capybara: The Breakthrough Tier
Capybara represents what Anthropic calls a “step change” rather than an incremental improvement. The leaked benchmarks show “dramatically higher scores” than Opus 4.6 in software coding, academic reasoning, and cybersecurity. The model creates “deep connective tissue between ideas and knowledge” — meaning it excels at tasks requiring cross-domain synthesis that no previous model could handle consistently.
The trade-offs are significant. Capybara is the slowest tier (expected), the most expensive to serve, and the most resource-intensive to run. Anthropic acknowledged it is “very expensive for us to serve, and will be very expensive for our customers to use.” This tier is not meant to replace Sonnet or Opus for everyday tasks — it exists for problems where nothing else is good enough.
Capybara vs Opus: Key Differences
The gap between Capybara and Opus is the largest between any two adjacent tiers in Anthropic’s lineup.
Performance Benchmarks
While Anthropic has not published exact scores, the leaked language is telling. The draft described Capybara’s improvement over Opus as “dramatic” — not moderate, not incremental. This word choice, across internal materials not intended for marketing, suggests a genuine performance leap. Specific areas of advancement include software coding, academic reasoning, and cybersecurity, where Mythos is described as “far ahead of any other AI model.”
Speed and Latency
Opus is already slower than Sonnet and Haiku. Capybara is expected to be slower still, given its larger model size and more intensive computation requirements. For interactive applications where response time matters, Opus or Sonnet will remain the practical choice. Capybara is better suited for batch processing, complex analysis tasks, and scenarios where accuracy outweighs speed.
Pricing Comparison
The full pricing picture for all four tiers, based on confirmed and estimated figures:
| Tier | Input (per 1M tokens) | Output (per 1M tokens) | Status |
|---|---|---|---|
| Haiku 4.5 | $0.80 | $4.00 | Available |
| Sonnet 4.6 | $3.00 | $15.00 | Available |
| Opus 4.6 | $5.00 | $25.00 | Available |
| Capybara (Mythos) | ~$10-25 (est.) | ~$50-125 (est.) | Restricted |
The jump from Opus to Capybara is expected to be the largest absolute price increase in the hierarchy. Community estimates range from 2x to 5x Opus pricing, based on Anthropic’s statements about the model being expensive to serve and the historical pattern of tier-to-tier pricing jumps.
Best Use Cases for Each Tier
Choosing the right tier depends on what you need. More expensive does not automatically mean better for your specific task.
When to Use Haiku
Use Haiku for high-volume, low-complexity tasks — text classification, basic summarization, content filtering, and any application where thousands of requests per minute matter more than nuanced understanding. Chatbots handling simple customer queries, data extraction from structured documents, and real-time content moderation all fit Haiku’s strengths.
When to Use Sonnet
Sonnet is the workhorse tier. Choose it for code generation, content creation, data analysis, and medium-complexity reasoning. Most production applications — from AI-assisted writing tools to automated code review — belong here. If you are building a product that needs to be both intelligent and cost-effective at scale, Sonnet is the default starting point.
When to Use Opus
Choose Opus when accuracy on complex tasks justifies the higher cost. Research analysis, architecture design, scientific reasoning, and any task where a wrong answer is more expensive than a slower or pricier response belong in Opus territory. Legal document analysis, medical research synthesis, and advanced mathematical problem-solving are Opus use cases.
When to Use Capybara
Capybara is for tasks that push beyond Opus’s limits. Based on Anthropic’s stated capabilities, primary Capybara use cases include automated vulnerability discovery in large codebases, enterprise-scale security auditing, massive codebase refactoring projects, and research tasks requiring cross-domain synthesis at a level no other model can match. If Opus gives you 90% of what you need and that last 10% matters enough to justify 2-5x the cost, Capybara is the answer.
Capybara Tier Pricing and Availability
Current API Pricing for All Tiers
As of March 2026, three of four tiers are publicly available through the Claude API:
Haiku 4.5 costs $0.80 per million input tokens and $4.00 per million output tokens. Sonnet 4.6 costs $3.00 and $15.00 respectively. Opus 4.6 costs $5.00 and $25.00. All three tiers support the same core features — tool use, vision, and extended context windows — differing primarily in intelligence and speed.
Estimated Capybara Pricing
Anthropic has not released official Capybara pricing. The leaked materials confirm high operational costs, and the company is reportedly working to improve efficiency before general availability. Based on the historical pricing pattern (roughly 2-4x between adjacent tiers) and Anthropic’s own statements, Capybara pricing will likely fall in the $10-25 range for input and $50-125 for output per million tokens.
This premium pricing means Capybara will not replace lower tiers for most applications. It is designed for high-value tasks where the cost per query is justified by the value of the output — security auditing, research breakthroughs, or critical infrastructure analysis.
Access and Release Timeline
Capybara is currently available only to a select group of early-access customers testing cybersecurity defense applications. Anthropic has not announced a public release date, stating that timing depends on safety evaluations rather than commercial schedules.
However, external factors suggest a likely window. Anthropic is reportedly considering an IPO in October 2026 with a valuation of $60 billion or more. Launching Capybara before the IPO would demonstrate product strength to potential investors. Industry analysts predict expanded testing in Q2-Q3 2026 and a probable public release in late 2026.
Questions About the Anthropic Capybara Tier
What is Anthropic’s Capybara tier?
Capybara is Anthropic’s new highest model tier, sitting above Opus in the Claude model hierarchy. It was revealed through a data leak on March 26, 2026. The first Capybara-tier model is Claude Mythos, described as a “step change” in AI capability.
How many model tiers does Anthropic have?
Anthropic now has four model tiers: Haiku (lightweight), Sonnet (balanced), Opus (flagship), and Capybara (breakthrough). Each tier trades off between speed, intelligence, and cost.
What is above Opus in Claude’s lineup?
Capybara sits above Opus. It is described as “larger and more intelligent than our Opus models” and achieves dramatically higher benchmark scores in coding, reasoning, and cybersecurity.
How much does Capybara cost?
No official pricing has been announced. Opus 4.6 costs $5/$25 per million tokens (input/output). Community estimates place Capybara at $10-25 input and $50-125 output per million tokens — a 2-5x premium over Opus.
When was Capybara announced?
Capybara was not formally announced. It was revealed through a CMS misconfiguration that exposed roughly 3,000 unpublished Anthropic assets on March 26, 2026. Fortune published the first report, and Anthropic confirmed the model’s existence the same day.
Is Capybara better than Opus?
Yes, in terms of raw capability. Capybara achieves dramatically higher benchmark scores than Opus 4.6 in coding, reasoning, and cybersecurity. The trade-off is slower response times and significantly higher cost. For many everyday tasks, Opus or Sonnet remains the practical choice.
What are the four Claude model tiers?
From lightest to most powerful: Haiku ($0.80/$4.00 per million tokens), Sonnet ($3.00/$15.00), Opus ($5.00/$25.00), and Capybara (pricing TBD, estimated 2-5x Opus). Each serves different use cases based on the balance of speed, intelligence, and cost needed.
What model is in the Capybara tier?
Claude Mythos is currently the only model in the Capybara tier. It was leaked on March 27, 2026, and is restricted to early-access cybersecurity testing. No additional Capybara-tier models have been announced.