Introduction

Claude Opus 4.7 was released by Anthropic on April 16, 2026. It is the most capable publicly available Claude model right now. It replaces Opus 4.6 as the flagship model and is available through claude.ai, the Anthropic API, Amazon Bedrock, Google Vertex AI, and Microsoft Azure Foundry.

The only model currently ahead of Opus 4.7 is Claude Mythos Preview. Anthropic has not released Mythos to the public due to safety concerns. So for anyone outside that restricted group, Opus 4.7 is the best Claude model you can access.

What Is Actually New in Opus 4.7

Self-Verification

Opus 4.7 devises ways to verify its own outputs before reporting back. This is a real behavioral change. Previous models would complete a task and hand back an answer. Opus 4.7 checks its own reasoning during the process. You will notice it catches logical errors in planning before they reach the final output.

High-Resolution Vision

Opus 4.7 is the first Claude model with high-resolution image support. Maximum image resolution has increased to 2576px at 3.75MP. The previous limit was 1568px at 1.15MP. This matters for analyzing dense documents, UI screenshots, charts, code diagrams, and design files.

Task Budgets

A task budget gives Claude a rough estimate of how many tokens to target for a full agentic loop. This includes thinking, tool calls, tool results, and final output. The model sees a running countdown and uses it to prioritize work and finish tasks gracefully as the budget is consumed.

You enable it by passing the beta header task-budgets-2026-03-13 and setting a task_budget inside output_config:

response = client.beta.messages.create(
    model="claude-opus-4-7",
    max_tokens=128000,
    output_config={
        "effort": "high",
        "task_budget": {"type": "tokens", "total": 128000},
    },
    messages=[
        {"role": "user", "content": "Review the codebase and propose a refactor plan."}
    ],
    betas=["task-budgets-2026-03-13"],
)

Adaptive Thinking

Extended thinking budgets are removed in Opus 4.7. Setting thinking: {"type": "enabled", "budget_tokens": N} will now return a 400 error. Adaptive thinking is the only thinking-on mode. In internal evaluations it reliably outperforms extended thinking.

Adaptive thinking is off by default. You enable it explicitly:

# Before (Opus 4.6)
thinking = {"type": "enabled", "budget_tokens": 32000}

# After (Opus 4.7)
thinking = {"type": "adaptive"}
output_config = {"effort": "high"}

New Effort Levels

Opus 4.7 introduces a new xhigh effort level between high and max. This gives users finer control over the tradeoff between reasoning depth and latency on hard problems. Anthropic recommends using high or xhigh when testing Opus 4.7 for coding and agentic tasks.

A New Tokenizer

Opus 4.7 uses a new tokenizer. It may use roughly 1x to 1.35x as many tokens when processing text compared to previous models. This is up to 35% more depending on content type. Code and structured data see the biggest token increases. The rate card has not changed, but your actual bill may go up.

Benchmark Results

On a 93-task coding benchmark, Opus 4.7 lifted resolution by 13% over Opus 4.6. This included four tasks that neither Opus 4.6 nor Sonnet 4.6 could solve.

Pricing

Pricing is the same as Opus 4.6. However, because of the new tokenizer, the same prompt will use more tokens than it did on Opus 4.6. A request that cost $0.10 on Opus 4.6 could cost anywhere from $0.10 to $0.135 on Opus 4.7 depending on content type. Prompt caching is the most reliable way to offset the tokenizer change.

Use Cases

Advanced Software Engineering

This is where Opus 4.7 is strongest. Users report being able to hand off their hardest coding work to Opus 4.7 with confidence. It handles complex long-running tasks with rigor and consistency and pays precise attention to instructions. If you have a multi-file refactor, a complex debugging session, or an architecture problem that Sonnet keeps getting wrong, Opus 4.7 is the right tool.

Long Agentic Pipelines

Opus 4.7 performs exceptionally well on long-horizon agentic work and scaled tool use. It is built for pipelines where the model must call multiple tools across many steps without human supervision in the middle. The task budget feature makes this even more reliable.

Data Integrity Work

The model correctly reports when data is missing instead of providing plausible-but-incorrect fallbacks. It resists dissonant-data traps that even Opus 4.6 falls for. This makes it a strong choice for financial analysis, scientific data processing, and any workflow where wrong answers with false confidence are expensive.

Visual Analysis

The 3.75MP image cap means you can now feed Opus 4.7 high-resolution screenshots, dense PDFs, architectural diagrams, and large UI mockups. It will read them with significantly more accuracy than older models.

Professional Document and Content Creation

Opus 4.7 is more tasteful and creative when completing professional tasks. It produces higher-quality interfaces, slides, and documents than its predecessor.

Who Should Use It

Opus 4.7 is the right model if you are working on tasks that are genuinely hard. This includes:

• Senior engineering work you would not trust a junior developer with
• Multi-step AI agents with tool calls
• Analytical work in domains where hallucinated data has real consequences
• Tasks where your current model keeps failing even after better prompting

For everyday tasks like drafting emails, answering general questions, or light coding, Sonnet 4.6 is faster and 40% cheaper per token. Use Opus 4.7 where the difficulty actually justifies it.

How to Access It

claude.ai — Select Opus 4.7 from the model picker. Available on Pro, Max, Team, and Enterprise plans.

API — Use the model string claude-opus-4-7. Available through Anthropic directly and through Amazon Bedrock, Google Vertex AI, and Microsoft Azure Foundry.

Claude Code — Opus 4.7 is now the default model for hard reasoning tasks.

GitHub Copilot — Available to Copilot Pro+, Business, and Enterprise users via the model picker.

Practical Tips

1. Turn On Adaptive Thinking for Hard Problems

Set thinking: {type: "adaptive"} and pair it with effort: "xhigh". Do not leave reasoning off when running complex engineering or multi-step analysis tasks. The xhigh effort level is new to Opus 4.7 and sits between high and max.

2. Set Task Budgets on Agentic Loops

If your agent runs for many steps, give it a token budget. The model will pace itself and finish cleanly rather than running out of context partway through a task.

3. Tell It to Be Concise

Opus 4.7 tends to be notably more verbose than its predecessor. If you do not need long explanations, say so explicitly in your prompt. The model follows instructions tightly so specificity helps.

4. Remove Temperature and Sampling Parameters

Setting temperature, top_p, or top_k to any non-default value will return a 400 error on Opus 4.7. The safest path is to omit these parameters entirely and use prompting to guide behavior instead.

5. Cache Your System Prompts

If you have a long system prompt or shared context across many requests, use prompt caching. You will get up to 90% off on cache reads. This helps absorb the cost increase from the new tokenizer.

6. Let It Check Its Own Work

Opus 4.7 is built to self-verify. Prompts like "before you answer, check your assumptions and correct any errors" push that behavior further. Do not rush it to a final answer on hard tasks.

7. Measure Your Real Token Costs Before Migrating

The new tokenizer means a potential 35% token increase on some content types. Run your existing Opus 4.6 prompts through count_tokens on Opus 4.7 before switching production traffic over.

Key Breaking Changes from Opus 4.6

• thinking: {"type": "enabled", "budget_tokens": N} returns a 400 error. Use thinking: {"type": "adaptive"} instead.
• temperature, top_p, and top_k now return a 400 error if set to non-default values.
• Thinking content is omitted from responses by default. Set "display": "summarized" to restore visible reasoning progress.
• The new tokenizer may increase token counts by up to 35% on the same prompts.

Opus 4.7 is the strongest publicly available Claude model today. It is built for autonomous, high-stakes, and technically demanding work. The self-verification behavior and task budget system are the two features that most change how you should design workflows around it.