The infrastructure Anthropic operates
Anthropic is the American company that builds Claude, currently operating five models in general availability: Fable 5, Opus 5, Sonnet 5 and Haiku 4.5, plus Mythos 5, which is invitation-only. The real capacity of this stack sits in coding and long-running agent execution, not image or video generation, which is entirely absent from this maker architecture.
- four models in general availability
- 1M token context window
- no image or video generation path
Last checked: This is a reference page. It is re-checked against the vendor sources and updated when a new version ships. Change log
The state of this maker's versions
- 7
- 4
- 2
- 2
- legacy 7
- current 4
- retired 2
- preview 2
These are the versions recorded in this reference, not the maker's whole catalogue.
Products
The product surface you actually operate, not the internal model names.
Model families
Four model lines, each positioned at a different point on the intelligence-versus-speed-and-price curve.
Versions
| Versions | Deployed | Status | Model families |
|---|---|---|---|
| Claude Fable 5.1 | 1 September 2026 | current | Claude Fable |
| Claude Mythos 5.1 | 1 September 2026 | preview | Claude Mythos |
| Claude Opus 5 | 24 July 2026 | current | Claude Opus |
| Claude Fable 5 | 9 June 2026 | legacy | Claude Fable |
| Claude Haiku 3.5 | retired | Claude Haiku | |
| Claude Haiku 4.5 | current | Claude Haiku | |
| Claude Mythos 5 | preview | Claude Mythos | |
| Claude Opus 4.1 | retired | Claude Opus | |
| Claude Opus 4.5 | legacy | Claude Opus | |
| Claude Opus 4.6 | legacy | Claude Opus | |
| Claude Opus 4.7 | legacy | Claude Opus | |
| Claude Opus 4.8 | legacy | Claude Opus | |
| Claude Sonnet 4.5 | legacy | Claude Sonnet | |
| Claude Sonnet 4.6 | legacy | Claude Sonnet | |
| Claude Sonnet 5 | current | Claude Sonnet |
Three layers that get collapsed into one
Most confusion about Anthropic comes from treating three distinct layers as a single unit. Claude is a product, the surface you sign in to and operate. Opus is a model family, a line with a fixed position on the intelligence-to-price curve. Opus 5 is a version, one specific set of weights with its own deployment date and price. When someone asks whether Claude outperforms GPT, it is unclear which layer they mean.
This page keeps the three layers separate, and every other page in this section is built on the same structure.
Where the operating envelope is defined, and where it stops
Anthropic is the only major lab with effectively no image, video or music generation path in its stack. We do not log that as a deficiency; it reads as an architectural choice — focus has stayed on text, code and agents. The practical consequence: route graphics or video workloads to Google or ByteDance, and keep Anthropic in the stack for code and analysis.
A documented detail that is easy to miss
From Opus 4.7 onward the tokenizer changed, and the same volume of text now produces roughly 30 percent more tokens. The nominal capacity of the million-token window is unchanged, but the actual volume of text that fits is lower than on earlier generations. If you have modeled a service cost on token throughput, migrating to a newer generation is not just a unit-price change — it shifts effective capacity too.
Documented strengths
- The most stable option currently available for coding workloads and multi-step agent execution
- A one-million-token context window across all four upper-tier public models
- Deployed simultaneously on Amazon Bedrock, Google Cloud and Microsoft Foundry, which for enterprises means a defined contract path and data region
- Relatively transparent documentation: knowledge cutoff and training data cutoff dates are logged separately
Known weaknesses and limits
- No image, video or music generation path in the architecture. The wrong pick for that workload.
- Its most expensive model, Fable 5 at $50 per million output tokens, costs several times the low end of the market.
- Mythos 5 is invitation-only with no self-serve signup path, so in practice it is not accessible to you.
- Blocked from Iran, and Iranian payment cards do not work either. No official, unobstructed route exists.
Technical verdict
If the workload is code, or an agent that has to run unattended for hours, Anthropic is the first option we evaluate today. If the workload is image or video, this stack has nothing for you and evaluating it wastes time.
Frequently raised questions, with documented answers
What is the difference between Fable 5 and Opus 5
Fable 5 is more expensive and slower, engineered for long-running agent execution, priced at $10 in and $50 out per million tokens. Opus 5 runs at half that price ($5 and $25) and is positioned for complex agentic coding and professional workloads. Opus 5 also carries a later knowledge cutoff.
Is Claude Code the same product as Claude
No. Claude is a conversational surface; Claude Code is a command-line tool that operates directly on a code repository. The underlying model layer can be shared, but the operating surface is not the same.
How much this tool is searched for
"anthropic" is searched 450,000 times a month in the United States and 5,400 times in Saudi Arabia. That is a 83x gap between two markets this site publishes in.
| Market | Term entered | Monthly search volume | Click cost | Ranking difficulty | User goal |
|---|---|---|---|---|---|
| Turkey | anthropic | 18,100 | $0.58 | 98 | Navigational |
| Saudi Arabia | anthropic | 5,400 | $1.17 | 79 | Navigational |
| United States | anthropic | 450,000 | $3.34 | 100 | Navigational |
Data from Semrush, TR, SA, US databases, retrieved 27 August 2026. Search volumes decay over time; read anything more than three months old as a trend line, not a fixed figure.