Claude Opus 5

anthropic/claude-opus-5

Claude Opus 5 is Anthropic's flagship and ranks #2 of 236 on the TryAii Score, 2 points behind Claude Fable 5.1. It tops the board outright on Chatbot Arena Elo (Document). It costs $5/$25 per million input/output tokens, though similar scores now sell for about $4.25 per million output tokens and it's reasonably quick at roughly 51-63 tokens/sec.

Prompt $/1M

$5.000

Completion $/1M

$25.000

Context

1,000,000 tok

Best provider

82 tok/s · Anthropic

Specs

Model id: anthropic/claude-opus-5
Display name: Claude Opus 5
Modality: text+image+file->text
Tokenizer: Claude
Instruct type: —
Context length: 1,000,000
Max completion: 128,000
Prompt $/1M: $5.000
Completion $/1M: $25.000
Free tier: no

Capabilities

Function callingno
JSON modeno
System promptno
Visionno
Moderatedyes

Benchmark scores

Each axis is one benchmark, normalized to the best model in the field.

No benchmark data for this model.

Speed by provider

Provider	tok/s	TTFT	Total latency
Anthropicfastest	82	3,232 ms	3,571 ms
Google	67	2,668 ms	3,121 ms
Artificial Analysis	53	20,694 ms	—
Claude Platform on AWS	51	4,384 ms	5,190 ms
Azure	51	2,400 ms	2,400 ms
Amazon Bedrock	47	4,579 ms	4,816 ms
anthropicfastest	—	—	—

Prompt caching by provider

Provider	Min tokens to cache	Notes
Anthropic	512 tokens	Fable 5 / Opus 5 tier — lowest Anthropic floor (Opus 5 halved from Opus 4.8's 1,024). Explicit cache_control breakpoints required; 5-min TTL refreshed free on every hit; 1h TTL opt-in at 2x write premium.
Google Vertex	512 tokens	Claude-on-Vertex, Fable 5 / Opus 5 tier: cache_control breakpoints exactly as on the direct API; per-model Anthropic tiers apply.
Amazon Bedrock	1,024 tokens	Fable 5 / Opus 5 on Bedrock: 1,024 assumed conservatively (512 on the direct API; Anthropic docs defer to AWS and the AWS side is unverified as of 2026-08).