Claude Opus 4.7: Features, Benchmarks & Pricing

Claude Opus 4.7 is Anthropic's most capable generally available AI model, released on April 16, 2026. Built for complex software engineering, long-horizon agentic tasks, and high-resolution vision work, it sits at the top of the Claude 4 family and is available via the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. This guide covers Claude Opus 4.7's specs, key features, benchmark performance, how it compares to its predecessor, and which similar Claude models you can access today on Chat Smith.

Claude Opus 4.7 is Anthropic's flagship model in the Claude 4 generation, succeeding Claude Opus 4.6. It is designed specifically for use cases where quality and reliability on hard tasks matter more than speed or cost: autonomous coding agents, multi-step document workflows, computer use, and enterprise knowledge pipelines. The model runs adaptive thinking by default and introduces several new developer-facing features including the xhigh effort level and task budgets.

Anthropic positions Opus 4.7 as the most capable model available to the general public, noting that Claude Mythos Preview remains more powerful overall but is not broadly available. For teams that need the highest quality accessible model for production agentic work, Opus 4.7 is the current ceiling.

Coding is the headline improvement in Opus 4.7. The model scores 64.3% on SWE-bench Pro, ahead of Claude Opus 4.6 (53.4%), GPT-5.4 (57.7%), and Gemini 3.1 Pro (54.2%). On CursorBench it reaches 70% versus 58% for Opus 4.6. Partner data from Rakuten shows it resolves 3x more production tasks than its predecessor on Rakuten-SWE-Bench. The model verifies its own outputs before reporting and uses fewer tool calls per task, making it more efficient in automated pipelines.

Opus 4.7 raises the maximum image resolution to 2,576px / 3.75MP — more than three times the pixel budget of Opus 4.6 at 1,568px / 1.15MP. On XBOW's visual-acuity benchmark, it scores 98.5% versus 54.5% for its predecessor. In practice, this means computer-use agents can read dense UI screenshots without downsampling, document analysis retains fine detail in charts and tables, and pixel coordinates map 1:1 with the image so there is no scale-factor math needed in computer use workflows.

A new xhigh ("extra high") effort level sits between high and max, giving developers finer control over the reasoning-vs-latency tradeoff. Anthropic recommends starting with high or xhigh for coding and agentic tasks. In Claude Code, xhigh is now the default effort level for all plans. At xhigh, the model approaches 75% on coding tasks while consuming fewer tokens than max effort — making it the practical sweet spot for most agentic workloads.

Task budgets (public beta) let developers give Claude a target token count for an entire agentic loop — including thinking, tool calls, tool results, and final output. The model sees a running countdown and self-moderates accordingly. This differs from max_tokens, which is a hard per-request cap the model doesn't know about. Task budgets help the model finish gracefully rather than cutting off mid-task, and are activated via the beta header task-budgets-2026-03-13. Minimum value: 20K tokens.

Opus 4.7 is more literal in how it executes instructions — where earlier models interpreted prompts loosely or skipped steps, Opus 4.7 follows them precisely. It is also better at writing to and reading from file-system-based memory stores, making it more reliable in long-running agents that use scratchpads or structured note files. Response length now calibrates to task complexity rather than defaulting to fixed verbosity.

Here is how Claude Opus 4.7 compares to its predecessor and key competitors on the benchmarks that matter most for production use:

Independent analysis from Artificial Analysis places Opus 4.7 at 54 on the Intelligence Index, well above average for reasoning models in its price tier. One notable data point from Hex: low-effort Opus 4.7 performs roughly equivalent to medium-effort Opus 4.6 — meaning you may achieve the same quality at lower compute cost if you calibrate effort levels correctly.

Understanding where Opus 4.7 sits relative to the broader Claude lineup and key competitors helps in making the right model choice for your use case:

For teams where cost matters, Claude Sonnet 4.6 delivers near-Opus performance at 40% lower input cost and 40% lower output cost. Anthropic's own guidance suggests Sonnet 4.6 is the right default for most production workloads, with Opus 4.7 reserved for the tasks where quality differences are measurable and worth paying for.

Opus 4.7 is strongest where quality on hard tasks justifies the premium:

Autonomous coding agents. Partners like Devin and Replit report the model works coherently for hours on complex multi-file codebases without losing context or requiring intervention. The 64.3% SWE-bench Pro score and 3x Rakuten task improvement make this the primary reason to choose Opus 4.7 over cheaper alternatives.

Computer use and visual automation. The 3.75MP resolution upgrade means computer-use agents can interact with dense UIs, read small text in screenshots, and handle high-DPI displays without image preprocessing. Combined with 1:1 pixel coordinate mapping, it simplifies GUI automation pipelines significantly.

Document and knowledge work. Opus 4.7 is state-of-the-art on GDPval-AA knowledge work benchmarks and handles complex multi-step document workflows with fewer tool errors. Finance, legal, and research pipelines that previously required Opus 4.6 will see measurable quality gains.

Enterprise agentic pipelines. Task budgets let teams set spending limits on agentic loops at the model level, which is especially useful for multi-turn agents where token costs can compound across many tool calls. Notion reported 14% improvement on complex workflows at fewer tokens and a third of the tool errors compared to Opus 4.6.

Claude Opus 4.7 is not currently available directly on Chat Smith, but Chat Smith gives you access to the closest available Claude models. Claude Sonnet 4.6 is the strongest Claude model on Chat Smith — it covers the same coding, reasoning, and long-context use cases that make Opus 4.7 compelling, at $3.00/1M input and $15.00/1M output, making it 40% cheaper per token. For Anthropic's latest Sonnet-tier release, Claude Sonnet 5 is also available and currently at promotional pricing through August 2026.

For high-volume or latency-sensitive tasks, Claude Haiku 4.5 is the budget Claude option on Chat Smith — fast and cost-efficient for classification, summarization, and routing tasks. Beyond Claude, Chat Smith also gives you access to GPT-5, Gemini 3.5 Flash, and Grok 4 — all from one interface.

Claude Opus 4.7 is a focused, practical upgrade that improves the areas where Opus was already strongest — coding, agentic reliability, and vision — without changing the price. The SWE-bench Pro jump from 53.4% to 64.3%, the 3.25x resolution increase, and new developer tools like xhigh effort and task budgets make it a clear upgrade for teams running complex production agents. If you want to explore the Claude model family today without API setup, Chat Smith gives you access to Claude Sonnet 4.6 and Claude Sonnet 5 alongside GPT, Gemini, Grok, and more — all from one interface. Browse the full model catalog on Chat Smith to find the right model for every task.