Claude Opus 5.5 is Anthropic’s current Opus model for coding, long-running agents and professional knowledge work. Its API model ID is claude-opus-5-5. Standard pricing is $4 per million input tokens and $20 per million output tokens, both 20% below Claude Opus 5.
Anthropic says typical token-billed workloads cost about 40% less because Opus 5.5 also uses fewer tokens per task. Treat that as a vendor workload estimate, not a guaranteed saving for every prompt or agent.
Pricing and cache costs
| Mode | Input | Cache read | Cache write | Output |
|---|---|---|---|---|
| Standard | $4.00 | $0.20 | $5.00 | $20.00 |
| Fast | $8.00 | — | — | $40.00 |
Prices are per million tokens. Fast mode is available in Claude Code and the Claude Platform, with Anthropic claiming up to 2.5× faster output. US-only inference costs 1.1× for input and output.
Cache reads fall from Opus 5’s $0.50/M to $0.20/M. That matters for coding agents that repeatedly reuse repository context, but cache economics still depend on whether requests actually hit a valid cache prefix.
Availability and migration boundaries
Opus 5.5 is available through the Claude Platform, AWS, Google Cloud and Microsoft Foundry. It supports zero data retention. Anthropic also makes it available to Claude Pro, Max, Team and Enterprise users, but app entitlements and API billing are separate.
Opus 5.5 launches with preserved thinking for API accounts created on or after August 31, 2026. Integrations must not edit prior assistant thinking blocks. Thinking also cannot be switched off, so test older message-history and proxy logic before changing the model ID.
Opus 5.5 in GitHub Copilot
GitHub is gradually rolling Opus 5.5 out to Copilot Pro+, Max, Business and Enterprise. It is selectable in VS Code, Visual Studio, Copilot CLI, the coding agent, Copilot App, github.com, GitHub Mobile, JetBrains, Xcode and Eclipse. Managed organizations can control it through model policies.
GitHub bills it at provider list pricing under Copilot usage-based billing. That does not mean every Copilot request maps directly to a single raw API call.
Should Opus 5 users migrate?
Opus 5.5 has lower list prices, cheaper cache reads and faster reported output, making it the logical evaluation target for new Opus deployments. Do not switch production solely from vendor benchmarks. Re-run coding, tool-use, structured-output, latency and safety evaluations, especially if your integration depends on mutable thinking history.
Anthropic reports more than 30% faster output and fewer steps on representative agent tasks. The practical metric is successful-task cost: include cache behavior, retries, tool calls and reviewer time.
Verdict
Claude Opus 5.5 is a meaningful cost-and-efficiency update rather than a packaging change. It is cheaper per token than Opus 5 and designed to finish agent work with fewer steps. Preserved-thinking compatibility is the main migration check before changing a production model ID.
For a cross-provider decision framework, compare Claude Opus 5.5 with GPT-6 Sol.