---
type: "post"
id: "what-a-task-costs-on-opus-5-5-0b85af"
slug: "what-a-task-costs-on-opus-5-5-0b85af"
title: "What a task costs on Opus 5.5"
description: "Anthropic’s Opus 5.5 guidance shifts the unit of cost from tokens purchased to agentic work completed."
retrieval_nugget: "Anthropic’s Opus 5.5 guidance shifts the unit of cost from tokens purchased to agentic work completed."
published_at: "2026-09-22"
updated_at: "2026-09-22"
record_date: "2026-09-22"
date_kind: "published_at"
topics: []
source_urls: ["https://claude.com/blog/what-a-task-costs-on-opus-5-5"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:what-a-task-costs-on-opus-5-5-0b85af"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:7f05d01296cdb3aac092d3c37b60eb4bdecf1031340ed22c9bd121af921f1372","reviewed_at":"2026-09-22T17:08:47.850Z","source_evidence_count":1,"verified_claim_count":3,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":3,"implication_count":2,"watch_condition_count":2,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"Anthropic’s Opus 5.5 guidance shifts the unit of cost from tokens purchased to agentic work completed.","observed_facts":[{"text":"The source says API input and output prices are 20% lower than Opus 5, while cache reads are 60% lower.","source_urls":["https://claude.com/blog/what-a-task-costs-on-opus-5-5"]},{"text":"It lists Opus 5.5 API prices of $4 per million input tokens, $20 per million output tokens, and $0.20 per million cache reads.","source_urls":["https://claude.com/blog/what-a-task-costs-on-opus-5-5"]},{"text":"Anthropic advises users to use `/usage` to inspect session input, output, and cache figures.","source_urls":["https://claude.com/blog/what-a-task-costs-on-opus-5-5"]}],"mechanism":"The mechanism is repeated context: every turn in a Claude Code task resends the accumulated conversation, while most previously seen material can be billed as a cache read.","why_now":"Why now is that lower cache-read pricing makes session design a more visible cost lever for long-running coding agents.","implications":["The first implication is that procurement and evaluation should compare cost per successfully completed task, not a model’s headline token rate.","Agent workflows that prevent false starts, support verification, and preserve cache continuity can have direct cost effects."],"evidence_boundary":"The evidence boundary is that Anthropic presents list prices and best-effort illustrations, says results vary by task, and excludes cache writes from its calculator.","watch_conditions":["The first watch condition is whether independent production measurements show that Opus 5.5 reduces turns or output tokens enough to extend savings beyond the published price cut.","Watch for actual cache-hit rates and retry rates across long-running Claude Code sessions."],"related_records":[],"new_branch_reason":"No prior New Runtime record was supplied; this official update begins a distinct record on task-level economics for Claude Code agents."}
routes: {"html":"https://newruntime.com/posts/what-a-task-costs-on-opus-5-5-0b85af/","markdown":"https://newruntime.com/posts/what-a-task-costs-on-opus-5-5-0b85af.md","json":"https://newruntime.com/posts/what-a-task-costs-on-opus-5-5-0b85af.json"}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
---

# What a task costs on Opus 5.5

## Retrieval answer

Anthropic’s Opus 5.5 guidance shifts the unit of cost from tokens purchased to agentic work completed.

Anthropic’s Opus 5.5 guidance shifts the unit of cost from tokens purchased to agentic work completed. The source says API input and output prices are 20% lower than Opus 5, while cache reads are 60% lower. It also lists Opus 5.5 API prices of $4 per million input tokens, $20 per million output tokens, and $0.20 per million cache reads.

The mechanism is repeated context: every turn in a Claude Code task resends the accumulated conversation, while most previously seen material can be billed as a cache read. The mechanism is repeated context: every turn in a Claude Code task resends the accumulated conversation, while most previously seen material can be billed as a cache read. This makes turn count, cache-hit rate, output—including billed thinking—and model choice jointly determine a task receipt. Anthropic’s illustrated 40-turn example processes roughly 2.8 million input tokens despite context peaking at 120K tokens; at a 90% cache-hit rate, it estimates input cost at about $1.62.

Why now is that lower cache-read pricing makes session design a more visible cost lever for long-running coding agents. Why now is that lower cache-read pricing makes session design a more visible cost lever for long-running coding agents. Teams should treat tests, builds, scoped tasks, and batching tool calls as economic controls when they reduce retries or unnecessary turns. The first implication is that procurement and evaluation should compare cost per successfully completed task, not a model’s headline token rate. The first implication is that procurement and evaluation should compare cost per successfully completed task, not a model’s headline token rate.

The evidence boundary is that Anthropic presents list prices and best-effort illustrations, says results vary by task, and excludes cache writes from its calculator. The evidence boundary is that Anthropic presents list prices and best-effort illustrations, says results vary by task, and excludes cache writes from its calculator. The first watch condition is whether independent production measurements show that Opus 5.5 reduces turns or output tokens enough to extend savings beyond the published price cut. The first watch condition is whether independent production measurements show that Opus 5.5 reduces turns or output tokens enough to extend savings beyond the published price cut.

Source: https://claude.com/blog/what-a-task-costs-on-opus-5-5

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)