Anthropic

Claude Opus 5 Pricing

Anthropic ยท Live

Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens on the Anthropic API. Cached input reads are billed at $0.5 per million tokens. It supports a 1,000,000-token context window and up to 128,000 output tokens per request. The model supports extended reasoning, image input, tool use, long context. Released July 24, 2026. Pricing verified November 4, 2026.

Claude Opus 5 API pricing

Claude Opus 5 API pricing is $5 per million input tokens and $25 per million output tokens, verified November 4, 2026.

Input (per 1M tokens)
$5
Output (per 1M tokens)
$25
Cached input (per 1M tokens)
$0.5
Cache write, 5m TTL
$6.25
Cache write, 1h TTL
$10
Batch discount
50%
Context window
1,000,000 tokens
Max output
128,000 tokens
Released
July 24, 2026
Estimate your Claude Opus 5 cost

Capabilities

Claude Opus 5 supports extended reasoning, image input, tool use, long context.

Extended reasoningImage inputTool useLong context

Claude Opus 5 API pricing has not changed since August 31, 2026.

Price-stable since August 31, 2026.

Frequently asked questions

Short answers to the most common questions about Claude Opus 5 pricing, limits, and capabilities.

Notes

Released July 24, 2026 at UNCHANGED Opus pricing ($5/$25 per MTok input/output) but with substantially higher capability. Anthropic positions it as "approaching Fable 5 intelligence at half the price." 1M-token context window. Benchmarks: IMO 2026 42/42 gold-medal performance (above the 29/42 cutoff), ARC-AGI-3 30.16% at high effort (~4x previous leaderboard best), Humanity's Last Exam 56.3% no tools / 64.7% with tools. More than doubles Opus 4.8 on agentic terminal coding benchmarks. Lowest misalignment score Anthropic has ever measured. This release effectively deprecates Opus 4.8 (same price, worse) and puts pricing pressure on Fable 5 ($10/$50 for marginal capability gain). Signals Anthropic competing on capability, not extracting rents โ€” likely a competitive response to GPT-5.6 Sol, Kimi K3, and Grok 4.5 crowding the frontier. [2026-08-31] Config backfill, no dollar price moved. Batch Discount % set to 50 (Batch API halves both input and output) and Max Output Tokens set to 128000 (synchronous limit; the Message Batches API allows up to 300K output via the output-300k-2026-03-24 beta header). Input $5 / output $25 / cache write 5m $6.25 / cache read $0.50 all re-verified unchanged. Sources: platform.claude.com pricing docs, finout.io/blog/anthropic-api-pricing, spheron.network Claude Opus 5 pricing. [2026-09-21] Cache Write 1h backfilled at $10.00 per MTok โ€” no dollar price moved, this field was previously empty. Sourced from platform.claude.com pricing coverage, which states both the rule ("a 1-hour cache write costs 2x base input") and the resulting Opus 5 figure ($10/M) in its per-model table. Consistent with the same 2x multiplier already stored on Haiku 4.5 ($1 input / $2 1h write) and Fable 5 ($10 input / $20 1h write). Caveat on the source: its prose separately mis-stated the 5-minute write as $10/M, contradicting its own table and the stored $6.25 (1.25x); the table figure was taken as correct because 1.25x/2x is Anthropic's documented structure and matches three other rows here. Input $5 / Output $25 / Cache Write 5m $6.25 / Cache Read $0.50 all re-verified unchanged. [2026-10-05] Data completion (no price moved): US-only Multiplier set to 1.1 (page: 1.1x on all token categories for Claude 4.6 and later via inference_geo "us"); Fast Mode Multiplier set to 2 (fast mode $10 / $50 vs standard $5 / $25) per https://platform.claude.com/docs/en/about-claude/pricing. Input $5 / Output $25 / cache read $0.50 re-verified unchanged.