Claude Design moved into a new phase on April 16, 2026, as Anthropic released Opus 4. 7 and positioned it as its latest production model. The update targets autonomous software engineering, longer agentic work, and higher-resolution visual processing. Pricing on the API stays unchanged at $5 per million input tokens and $25 per million output tokens.
Claude Design gains in coding, vision, and long-task handling
The most striking result in Claude Design is the jump on SWE-bench Pro, the benchmark tied to agentic coding. Opus 4. 7 reaches 64. 3 percent, ahead of Opus 4. 6 at 53. 4 percent, GPT-5. 4 at 57. 7 percent, and Gemini 3. 1 Pro at 54. 2 percent. On SWE-bench Verified, the score rises to 87. 6 percent.
Anthropic also says the model reaches 94. 2 percent on GPQA Diamond, a doctoral-level reasoning benchmark, putting it close to GPT-5. 4 Pro at 94. 4 percent. The company is presenting Claude Design as a model built for long, complex tasks with limited supervision, including multi-session projects that must stay coherent from start to finish.
New visual support and a finer API control point
A second shift in Claude Design is image handling. Opus 4. 7 processes images at a resolution three times higher than Opus 4. 6, which Anthropic says should improve interfaces, slides, and generated documents. The related release note also points to a maximum image size of 2, 576 pixels on the long side.
On the API, Anthropic adds a new effort level called “xhigh, ” placed between “high” and “max. ” That gives developers a more precise way to balance reasoning depth and latency. The model also keeps the same token pricing as Opus 4. 6, a detail that matters for teams watching cost as closely as capability.
Claude Design and the competition for professional workflows
The new release lands in a crowded field. Anthropic places Opus 4. 7 against GPT-5. 4 and Gemini 3. 1 Pro across several evaluations, while also stressing that Claude Mythos Preview remains reserved for a limited set of cybersecurity partners in the Project Glasswing program. In that setup, Anthropic is effectively running two tracks: Opus for commercial deployment and Mythos for the frontier.
Anthropic says the model also shows stronger literal instruction-following, which may force developers to revisit prompts that worked more loosely on Opus 4. 6. The same release highlights filesystem-based memory across sessions, meant to preserve notes and reduce the need to rebuild context.
Immediate reaction from the benchmark picture
Artificial Analysis identified GDPval-AA, the evaluation Anthropic cites for economically valuable knowledge work across finance, law, and related fields. Anthropic says Opus 4. 7 leads that benchmark, alongside a state-of-the-art result on Finance Agent, reinforcing the company’s push beyond coding alone. In practical terms, Claude Design is being framed as a tool for professional workflows that mix analysis, documents, and long-running execution.
What happens next for Claude Design
Claude Design is already available to Pro, Max, Team, and Enterprise subscribers, and through the API, Amazon Bedrock, Vertex AI, and Microsoft Foundry. The next test will be how quickly developers adapt prompts, workflows, and review processes to the model’s more literal instruction handling and higher-resolution visual output. For teams watching the pace of model releases, Claude Design now sits at the center of Anthropic’s most aggressive production push yet.
