Claude Opus 4.7: Anthropic's Most Capable Model with xhigh Effort and Adaptive Thinking

Claude Opus 4.7: Anthropic Reclaims the Frontier

On April 16, 2026, Anthropic released Claude Opus 4.7, its most capable generally available model to date. The release arrives with a clear positioning: Opus 4.7 is Anthropic's answer to the past few months of frontier-model competition, posting headline benchmark results and — more importantly — delivering measurable real-world gains in the kind of long-horizon, tool-using, agentic coding tasks that the Claude family has become known for.

Benchmark Results

Opus 4.7 posts strong results across the benchmarks that matter for developer work:

• SWE-bench Verified: 87.6% — the highest score of any generally available model at launch
• SWE-bench Pro: 64.3% — the hardest publicly available real-world coding benchmark, where Opus 4.7 outperforms OpenAI's GPT-5.4 and every other major frontier model
• CursorBench: 70% (vs 58% for Opus 4.6) — a 12-point jump on a benchmark calibrated to real IDE tasks
• General coding improvements: Anthropic reports a ~13% improvement over Opus 4.6 on internal coding evaluations

The gap over Opus 4.6 is not subtle. On coding-heavy benchmarks the improvement is roughly one full generational step, and Anthropic has been careful to note that the model was designed specifically for long-horizon agentic work — the category where previous models tended to lose context or compound errors.

What's New in the Model

A New Effort Tier: xhigh

Opus 4.7 introduces a new xhigh effort level that sits between high and max. It is designed as the new sweet spot for agentic work: substantially more reasoning than high, with meaningfully lower latency and cost than max. Anthropic recommends xhigh as the default for most coding and agentic use cases and reserves max for genuinely hard problems where the additional cycles justify the overhead.

Adaptive Thinking Replaces Fixed Budgets

Fixed thinking budgets are gone. Opus 4.7 uses adaptive thinking, deciding dynamically at each step whether extended reasoning adds value. The model is faster on simple queries and more deliberate on complex ones — without requiring any configuration from the developer. Anthropic notes it is also less prone to overthinking than Opus 4.6, which previously introduced unnecessary latency and cost in agentic chains.

High-Resolution Image Support Tripled

Image support ceiling jumps from 1,568 pixels to 2,576 pixels — roughly 3.75 megapixels, more than three times the prior ceiling. This is a meaningful unlock for multimodal workflows: design review, screenshot-driven debugging, chart interpretation, and UI analysis all benefit directly.

Long-Horizon Agentic Reliability

Bolt's evaluation reports Opus 4.7 worked on a previously AI-resistant coding problem for 7.5 hours — with individual reasoning blocks exceeding one hour — and arrived at a legitimate solution without looping or regressing. Context compression is another area of meaningful progress: Opus 4.7 compressed its context three to four times on a single task and continued executing against the original plan each time, a scenario where prior models typically drifted.

Reduced Tool and Subagent Usage by Default

Opus 4.7 reasons more internally and spawns parallel subagents more conservatively than Opus 4.6. Teams that relied on aggressive tool invocations or parallel delegation will need to prompt for that behavior explicitly.

Cybersecurity Safeguards

The release ships with automated cybersecurity safeguard checks built in. Anthropic has also expanded its evaluation suite around misuse resistance, and Opus 4.7 is positioned as a lower-risk model than the unreleased "Mythos" system referenced internally.

Availability and Pricing

Opus 4.7 is available on day one across every major distribution surface:

• Claude API — direct access for developers
• Amazon Bedrock and Google Vertex AI — enterprise cloud integrations
• Claude Code — now the default model, shipped in v2.1.111
• GitHub Copilot — replacing Opus 4.5 and 4.6 in the model picker for Pro+ subscribers
• Bolt, Codex, Mistral Vibe, Cursor — integrated via downstream tools

Pricing remains unchanged from Opus 4.6: $5 per million input tokens and $25 per million output tokens. Given the meaningful capability gains and stable pricing, Anthropic is clearly positioning this as a pure upgrade rather than a premium tier.

Business Context

The release coincides with Anthropic reaching an annual revenue run-rate of $30 billion, driven heavily by enterprise adoption and the rapid growth of Claude Code. Opus 4.7 is Anthropic's most visible move to reclaim the frontier narrative heading into the next release cycle, and the simultaneous availability across every major developer surface underscores how central the coding use case has become to Anthropic's strategy.