Anthropic

Anthropic — Claude models, built around long agentic runs and tool use. Anthropic sells the Claude family through a first-party API, plus Bedrock, Vertex and Foundry. The current line-up spans Fable 5 at the top, Opus 4.8 as the workhorse and Sonnet 5 for volume — each with a one-million-token input window and a 128K output cap — plus Haiku 4.5 for latency, which is a 200K-window model rather than a million-token one. The platform's distinguishing features are graduated effort control, prompt caching with explicit breakpoints, and a managed agent surface — the things a long tool-calling loop actually needs. Founded 2021. Late-stage private.

We track it in 1 comparison — LLM APIs — so every claim below is a cell in a table you can open and check rather than an impression. Across those rosters it sits against 17 other tools, and what follows is where it visibly separates from them.

Where it wins.

• Effort control — Yes, which only 4 of 18 tools here manage.
• Context window — 1,000,000 tokens, tied for best across 4 of 10 tools.
• MCP support — Yes, which only 2 of 18 tools here manage.
• Computer use — Yes, which only 2 of 18 tools here manage.

Each of those is ranked against the whole roster on its category page, not against a hand-picked subset, so a first place here means first of everything we list.

Where it gives ground.

• $/M output — $50 /M tok, the weakest of the 8 here. DeepSeek records $0.42 /M tok.
• $/M input — $10 /M tok, the weakest of the 8 here. DeepSeek records $0.28 /M tok.
• $/M cache read — $1 /M tok, the weakest of the 5 here. DeepSeek records $0.028 /M tok.
• Open weights — No. Meta Llama records Yes.

None of these disqualify it on their own. They are the fields to check against your own requirements before you commit, because they are the ones where a competitor genuinely does better.

Provenance. 25 of 27 tracked fields carry a value for Anthropic, and 17 of those cite a document you can open. Last verified 2026-06-24. Every figure keeps its own provenance — measured by us, claimed by the vendor, inferred, or community-reported — and we would rather print a dash than a guess.

Its nearest neighbour in our data is Qwen. Anthropic is ahead on Computer use (Yes against No) and Context window (1,000,000 tokens against 262,144 tokens). Qwen takes $/M output ($6 /M tok against $50 /M tok) and $/M input ($1.2 /M tok against $10 /M tok). That pattern repeats across the rest of the roster — see Anthropic alternatives for the other rivals, each compared the same way.

Founded

2021

Funding

Late-stage private

Fields we track

25 of 27

Last verified

2026-06-24

The best platform here for anything that runs a tool loop for more than a few turns — effort control, explicit cache breakpoints and MCP are designed for that shape of work rather than retrofitted. The continuity story regressed with the current generation: dropping dated snapshot IDs means you pin to a published retirement date, not to an immutable model. One flagship-specific trap worth wiring for on day one: Fable 5 can decline a request outright, returning HTTP 200 with stop_reason 'refusal' and no usable content, so a client that reads the first content block unconditionally breaks rather than errors. Anthropic ships a fallbacks parameter that re-runs the request on another model; use it, or handle the stop reason yourself.

$/M output

$50 /M tok8th of 8

Vendor-claimedverified 2026-06-24source

$/M input

$10 /M tok8th of 8

Vendor-claimedverified 2026-06-24source

Notice period

180 days1st of 2

Inferred

$/M cache read

$1 /M tok5th of 5

Inferredverified 2026-06-24source

Effort control

●Yes1st of 18

Vendor-claimedverified 2026-06-24source

Context window

1,000,000 tokens1st of 10

Vendor-claimedverified 2026-06-24source

MCP support

●Yes1st of 18

Vendor-claimedverified 2026-06-24source

Open weights

○No18th of 18

Inferredverified 2026-06-24

Anthropic is the right call in these situations, each one drawn from a field we actually record:

• Agent harnesses that call tools for minutes at a time.
• Code generation and review where first-try correctness pays for the token price.
• Teams that want cost tuned per route via graduated effort levels.
• Effort control is your binding constraint. Anthropic records Yes, which only 4 of the 18 tools in the LLM APIs roster do. We define that field as a request-level parameter that trades reasoning depth against cost and latency;.
• Context window is your binding constraint. Anthropic records 1,000,000 tokens, the best figure in the LLM APIs roster. We define that field as maximum INPUT tokens the flagship model accepts in a single request, at the standard (non-surcharged) tier.
• MCP support is your binding constraint. Anthropic records Yes, which only 2 of the 18 tools in the LLM APIs roster do.
• Computer use is your binding constraint. Anthropic records Yes, which only 2 of the 18 tools in the LLM APIs roster do.

Reach for something else when any of the following is a requirement rather than a nice-to-have:

• $/M output. Anthropic records $50 /M tok, the weakest of the 8 tools in the LLM APIs roster. DeepSeek records $0.42 /M tok on the same field. If that is a hard requirement rather than a preference, start elsewhere.
• $/M input. Anthropic records $10 /M tok, the weakest of the 8 tools in the LLM APIs roster. DeepSeek records $0.28 /M tok on the same field. If that is a hard requirement rather than a preference, start elsewhere.
• $/M cache read. Anthropic records $1 /M tok, the weakest of the 5 tools in the LLM APIs roster. DeepSeek records $0.028 /M tok on the same field. If that is a hard requirement rather than a preference, start elsewhere.
• Open weights. Anthropic records No on the LLM APIs table. Meta Llama records Yes on the same field. If that is a hard requirement rather than a preference, start elsewhere.

We publish this block because a comparison that only lists what a tool is good at is marketing. Every figure above sits on the same page as its source, and the field definitions are on the category tables if you want to check how we measured them.

What we don't know. 2 of the 27 fields we track for Anthropic are still blank: TTFT p50 and Output tok/s. Those render as dashes rather than as zeroes or assumptions, because an empty cell and a bad cell are not the same thing and only one of them is honest. If you know any of these figures and can point at a document, tell us.