Claude Opus 5: Features, Pricing, Benchmarks and Access
Anthropic released Claude Opus 5 on July 24, 2026, as its new high-capability model for complex coding, long-running agents, research, document work, and enterprise tasks. It offers a one-million-token context window, output of up to 128,000 tokens, adaptive thinking by default, and the same standard API pricing as Claude Opus 4.8.
The model is not designed to replace every cheaper Claude option. Its strongest value appears when a task requires deeper reasoning, several tool calls, work across a large codebase, or careful verification over a long session. For everyday questions and high-volume workloads, Claude Sonnet 5 or Haiku may still be more economical.
This guide explains the confirmed Claude Opus 5 features, API pricing, availability, benchmark claims, use cases, limitations, and differences from Sonnet 5 and Opus 4.8.
Quick answer: Claude Opus 5 is Anthropic’s advanced model for complex agentic coding and professional work. It supports a 1M-token context window and up to 128K output tokens. Standard API pricing is $5 per million input tokens and $25 per million output tokens. Its API model ID is
claude-opus-5.
What Is Claude Opus 5?
Claude Opus 5 is the latest model in Anthropic’s Opus family. Opus models are positioned above Sonnet and Haiku for tasks where accuracy, reasoning depth, autonomy, and careful judgment matter more than minimum cost or maximum speed.
Anthropic describes Opus 5 as a major improvement over Opus 4.8 rather than a small update. Its official documentation highlights gains in:
- Deep, multi-step reasoning
- Agentic coding and long-horizon tasks
- Code review and bug detection
- Complex document and office work
- Visual understanding
- Long-context instruction following
- Multi-agent coordination
- Efficient performance at lower effort settings
The practical difference is that Opus 5 is intended to maintain a goal across a larger workflow. Instead of only generating a code snippet or summarizing a file, it can plan, use tools, inspect results, correct mistakes, and continue until a broader task is complete.
This is part of the wider shift from simple chatbots toward AI tools that can assist with serious development work.
Key Claude Opus 5 Features
One-Million-Token Context Window
Claude Opus 5 supports a 1M-token context window as both its default and maximum context size on the API. Anthropic also says the 1M context window is available when chatting with Opus 5 on paid Claude plans.
A larger context window can hold a large codebase, hundreds of document pages, extensive research, or a long project history. It does not guarantee perfect recall, so users should still organize files and label important instructions clearly.
Up to 128K Output Tokens
The model can produce up to 128,000 output tokens through supported developer workflows, which is useful for substantial reports, multi-file code changes, and structured deliverables. Unnecessarily long output still increases cost and review time.
Adaptive Thinking by Default
Thinking is enabled by default in Claude Opus 5. The model decides when deeper reasoning is useful and how much thinking a particular turn needs.
Developers control the tradeoff between capability, latency, and token usage through the effort parameter: low, medium, high, xhigh, or max.
Anthropic recommends starting with the default setting and increasing effort only when the task genuinely requires it.
Better Agentic Coding
Claude Opus 5 is built for work that extends beyond code completion. Anthropic says it is better at:
- Navigating large repositories
- Implementing multi-file features
- Completing larger refactors
- Finding root causes instead of patching symptoms
- Reviewing code with fewer false positives
- Using terminals, browsers, and other tools
- Testing and verifying its own work
These capabilities make it relevant to Claude Code, IDE integrations, and software agents. Developers should still review changes, run tests, and protect production credentials.
More Proactive Verification
Anthropic’s migration guidance says Opus 5 checks its own work more often without being explicitly told to do so. It may also provide more progress narration and delegate to subagents more readily in multi-agent systems.
This behavior can reduce the need for repetitive prompt instructions such as “verify everything at the end.” Anthropic warns that keeping old verification instructions may cause over-verification, increasing latency and token consumption.
Stronger Vision and Document Work
Opus 5 can analyze charts, documents, diagrams, screenshots, and visual interfaces. Anthropic specifically highlights frontend replication, spreadsheet creation, formula-heavy workbooks, presentation generation, and document editing.
Visual tasks benefit from clear references, tools for inspecting the result, and a defined acceptance checklist.
Claude Opus 5 Pricing
Claude Opus 5 keeps the same base API rates as Opus 4.8:
| Usage type | Price per million tokens |
| Standard input | $5.00 |
| Standard output | $25.00 |
| 5-minute prompt cache write | $6.25 |
| Prompt cache hit or refresh | $0.50 |
| Batch API input | $2.50 |
| Batch API output | $12.50 |
| Fast mode input | $10.00 |
| Fast mode output | $50.00 |
Fast mode is a research-preview option that Anthropic says runs around 2.5 times faster than standard mode. It costs twice the standard rate and is currently limited to Anthropic’s first-party Claude API and supported usage-credit workflows in Claude Code. It is not available through Amazon Bedrock, Google Cloud, or Microsoft Foundry.
The Batch API provides a 50% discount for asynchronous workloads that do not need an immediate response.
Example API Cost
Suppose one standard Opus 5 request uses:
- 100,000 input tokens
- 10,000 output tokens
The approximate model cost would be:
- Input: 0.1 × $5 = $0.50
- Output: 0.01 × $25 = $0.25
- Estimated total: $0.75
This example excludes server-side tools, data-residency modifiers, retries, cache writes, and any other platform charges.
How to Access Claude Opus 5
Claude App
Anthropic says Opus 5 is the new default model on Claude Max and the strongest model available on Claude Pro. Availability and usage limits depend on the user’s plan, location, conversation size, and current product rules.
Users on paid plans can select an available model from the Claude model picker. If Opus 5 does not appear, check the account plan, app version, and regional availability.
Claude Code
Claude Code lets developers use Claude through a terminal or supported IDE. Opus 5 is best reserved for hard debugging, architecture decisions, large refactors, or planning that benefits from deeper reasoning. Anthropic’s own usage guidance suggests Sonnet for routine coding and Opus for the most difficult work.
Claude API
Developers can call the model using:
claude-opus-5
Opus 5 is available to Claude API customers. API billing is separate from a normal Claude Pro or Max subscription unless a particular usage-credit workflow explicitly states otherwise.
Cloud Platforms
Anthropic lists Claude Opus 5 on:
- Amazon Bedrock
- Google Cloud
- Microsoft Foundry
- Anthropic’s first-party Claude API
Fast mode remains limited to the first-party API.
Claude Opus 5 Benchmarks: What Anthropic Reports
Benchmark numbers should be interpreted carefully. Test setup, tool access, reasoning effort, cost limits, and task selection can materially affect results. Independent testing is still important, particularly because Opus 5 is newly released.
In its July 24 announcement, Anthropic reported that:
- Opus 5 more than doubled Opus 4.8’s performance on Frontier-Bench v0.1 while costing less per task.
- At maximum effort, it came within 0.5% of Claude Fable 5’s peak CursorBench 3.2 result at about half the cost per task.
- Its ARC-AGI 3 result was three times the next-best model in Anthropic’s comparison.
- Its Zapier AutomationBench pass rate was around 1.5 times the next-best model at the same cost per task.
- It outperformed the compared models on OSWorld 2.0 at a given cost.
These are vendor-reported results, not conclusions from our own hands-on testing. They indicate where Anthropic expects the model to be strongest: software engineering, computer use, automation, and difficult knowledge work.
For a real business decision, test Opus 5 on your own tasks and compare completion accuracy, tokens, retries, time, human review effort, and cost per successful result.
A model with a higher per-token price can still be cheaper overall if it solves the task in fewer attempts. The opposite can also be true for simple work.
Claude Opus 5 vs Sonnet 5 vs Fable 5
| Feature | Claude Opus 5 | Claude Sonnet 5 | Claude Fable 5 |
| Best fit | Complex coding, enterprise work, deep reasoning | Everyday coding, agents and balanced workloads | Highest-capability long-running agents |
| Context window | 1M tokens | 1M tokens | 1M tokens |
| Maximum output | 128K tokens | 128K tokens | 128K tokens |
| Standard input price | $5/MTok | $3/MTok after Aug. 31, 2026 | $10/MTok |
| Standard output price | $25/MTok | $15/MTok after Aug. 31, 2026 | $50/MTok |
| Current Sonnet promotion | Not applicable | $2 input/$10 output through Aug. 31, 2026 | Not applicable |
| Main advantage | Strong capability at half Fable’s base token price | Better speed-and-cost balance | Anthropic’s highest widely released capability |
Choose Sonnet 5 for routine coding, drafting, everyday analysis, and high-volume tasks where speed and budget matter.
Choose Opus 5 when the job involves difficult debugging, cross-cutting changes, ambiguous requirements, advanced research, complex documents, or long multi-step workflows.
Choose Fable 5 only when the highest available capability is worth paying twice Opus 5’s base token price.
For many teams, a mixed workflow is more efficient: use Opus 5 to reason about the architecture or create a detailed plan, then use Sonnet 5 for more mechanical execution.
Claude Opus 5 vs Opus 4.8
Claude Opus 5 is priced identically to Opus 4.8, which makes it an attractive upgrade for new work. Anthropic describes it as a drop-in upgrade, but developers should review two behavior changes:
- Thinking is enabled by default.
- Disabling thinking while using
xhighormaxeffort produces an API error.
Opus 5 may also produce longer deliverables, narrate more progress during agentic work, verify results more often, and use subagents more readily.
Developers migrating an application should change the model ID from claude-opus-4-8 to claude-opus-5, then run regression tests for cost, latency, output style, tool calls, and safety fallback behavior.
Opus 4.8 remains available, so teams do not need to migrate every production workflow immediately.
Best Use Cases for Claude Opus 5
Large Software Projects
Opus 5 suits repository-wide refactoring, root-cause debugging, pull-request review, architecture planning, and long tool-using sessions.
Research and Analysis
The model can compare large document collections and produce structured analysis. Human review remains essential for medical, legal, financial, or scientific decisions.
Business Automation
Opus 5 can support workflows involving spreadsheets, account analysis, internal tools, document generation, and multi-step operations. It may be useful when simpler AI automation workflows fail because the task requires judgment across several steps.
Complex Documents and Presentations
Anthropic highlights stronger performance on formula-heavy spreadsheets, slide decks, and professional documents. This makes the model relevant to consultants, analysts, operations teams, and enterprise knowledge workers.
Vision-Heavy Workflows
It can analyze screenshots, charts, diagrams, and interface references. Developers can use this capability for frontend recreation, visual QA, document extraction, or applications that combine images and text.
For basic writing, scheduling, summaries, and other routine productivity work, less expensive AI tools can save hours without requiring a frontier model.
Limitations and Practical Concerns
Claude Opus 5 is not automatically the right model for every task.
- Higher cost: It costs more than Sonnet 5 and Haiku per token.
- Longer responses: Its default behavior may produce more detailed deliverables.
- Variable latency: Higher effort settings and tool-heavy tasks take more time.
- Usage limits: Consumer plans have session limits, and large contexts consume quota faster.
- No guarantee of correctness: It can still make factual, coding, or reasoning errors.
- Benchmark dependence: Vendor benchmarks may not represent a specific company’s workflow.
- Preview features: Fast mode and some platform features remain in research preview or beta.
- Migration differences: Default thinking and effort behavior can affect existing API applications.
Use access controls, human approvals, tests, backups, and spending limits for any workflow that can change production systems or sensitive data.
Is Claude Opus 5 Worth It?
Claude Opus 5 is worth testing when your work is difficult enough that cheaper models require repeated correction or fail to finish the full task. Its unchanged price relative to Opus 4.8 makes the decision particularly straightforward for teams already paying for the previous Opus model.
Sonnet 5 remains the more practical default for ordinary work. Opus 5 should be used deliberately for tasks where deeper reasoning and better long-horizon performance can justify the additional cost.
The best evaluation is not “Which model has the highest benchmark score?” It is “Which model produces a correct, usable result at the lowest total cost and review effort for this task?”
Frequently Asked Questions
When was Claude Opus 5 released?
Anthropic released Claude Opus 5 on July 24, 2026.
How much does Claude Opus 5 cost?
Standard Claude API pricing is $5 per million input tokens and $25 per million output tokens. Fast mode costs $10 per million input tokens and $50 per million output tokens.
What is the Claude Opus 5 context window?
Claude Opus 5 supports a one-million-token context window as its default and maximum API context size. It can generate up to 128,000 output tokens.
What is the Claude Opus 5 API model ID?
The API model ID is claude-opus-5.
Is Claude Opus 5 better than Sonnet 5?
Opus 5 is designed for more complex reasoning, coding, research, and long-running work. Sonnet 5 offers a better speed-and-cost balance for everyday and high-volume tasks. The better choice depends on the workload.
Is Claude Opus 5 available in Claude Code?
Yes. Opus 5 is available in Claude Code, subject to the user’s plan and usage limits.
Does Claude Opus 5 replace Opus 4.8?
It is the recommended upgrade for many Opus 4.8 workloads, but Opus 4.8 remains available. Developers should test the new default thinking behavior and other migration differences before changing production systems.
Final Thoughts
Claude Opus 5 is a significant new option for developers and professional teams that need more than a fast chatbot. Its one-million-token context window, long output capacity, adaptive thinking, stronger verification, and improved agentic behavior are designed for demanding workflows.
Its $5 input and $25 output API pricing is unchanged from Opus 4.8, while Anthropic reports meaningful performance improvements. That combination makes Opus 5 an obvious model to evaluate for existing Opus users.
Do not use it everywhere by default. Start with Sonnet for normal work, move to Opus 5 when complexity rises, and compare the total cost of successful completion rather than the price of a single token.
