Anthropic

Claude models, built around long agentic runs and tool use

anthropic.com platform.claude.com/docs

Overall in llm apis

12of 18 · score 61

Data coverage

93%

25 of 27 fields sourced

Best for

•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

Strong on

+Effort control Yes 1/18
+MCP support Yes 1/18
+Computer use Yes 1/18
+Notice period 180 days 1/7

Weak on

−$/M output $50/M tok 13/13
Better: Fireworks AI: $0.05/M tok
−$/M input $10/M tok 13/13
Better: Groq: $0.05/M tok
−$/M cache read $1/M tok 12/12
Better: DeepSeek: $0.0028/M tok
−Open weights No 18/18
Better: Meta Llama: Yes

vs Google Gemini

•Ahead on MCP support: Yes vs Partial
•Ahead on Computer use: Yes vs Partial
•Behind on $/M output: $50/M tok vs $12/M tok
•Behind on $/M input: $10/M tok vs $2/M tok

How Anthropic scores among 18 llm apis25 of 27 fields known

Capability7

Flagship model

Claude Fable 5 (claude-fable-5)

platform.claude.com · 2026-06-24

Context window1st

1,000,000 tokens

1/14

platform.claude.com · 2026-06-24

Max output

128,000 tokens

4/10

Leader: xAI · Unlimited

platform.claude.com · 2026-06-24

Image input1st

Yes

1/18

platform.claude.com · 2026-06-24

Audio in/out

11/18

Leader: OpenAI · Partial

platform.claude.com · 2026-06-24

Open weights

18/18

Leader: Meta Llama · Yes

· 2026-06-24

OpenAI-compat API

Partial

17/18

Leader: OpenAI · Yes

· 2026-06-24

Performance2

TTFT p50

—

Output tok/s

—

Pricing4

$/M input

$10/M tok

13/13

Leader: Groq · $0.05/M tok

platform.claude.com · 2026-11-06

$/M output

$50/M tok

13/13

Leader: Fireworks AI · $0.05/M tok

platform.claude.com · 2026-11-06

$/M cache read

$1/M tok

12/12

Leader: DeepSeek · $0.0028/M tok

platform.claude.com · 2026-11-06

Batch discount1st

50%

1/14

platform.claude.com · 2026-06-24

Agentic6

Tool use1st

Yes

1/18

platform.claude.com · 2026-06-24

Schema output1st

Yes

1/18

platform.claude.com · 2026-06-24

Effort control1st

Yes

1/18

platform.claude.com · 2026-06-24

Computer use1st

Yes

1/18

platform.claude.com · 2026-06-24

MCP support1st

Yes

1/18

platform.claude.com · 2026-06-24

Cache TTL

5 min default, 1 h option; automatic or explicit breakpoints

platform.claude.com · 2026-06-24

Governance & continuity6

Continuity policy

Published deprecation page lists a retirement date per model, typically months ahead, Opus 3 went on 2026-01-05, Sonnet 3.7 and Haiku 3.5 on 2026-02-19. Older models carried dated snapshot IDs you could pin, but the current generation ships as bare aliases with no dated ID behind them, so the retirement date is your only guarantee.

platform.claude.com · 2026-06-24

Notice period1st

180 days

1/7

Pinnable versions

Partial

14/18

Leader: OpenAI · Yes

platform.claude.com · 2026-06-24

Zero retention

Partial

7/18