Anthropic made Claude Opus 4.7 generally available on April 22, 2026. The model is accessible via the Claude API, and GitHub Copilot Pro+, Business, and Enterprise users can select it directly in the model picker, according to GitHub’s changelog.
The confirmed capability profile includes a 1M token context window, 128K maximum output tokens, adaptive thinking, and what Anthropic describes as enhanced visual analysis for professional document workflows. Anthropic’s technical report accompanies the release as arXiv:2604.09881, a vendor-authored document, not an independent evaluation.
On benchmarks, Anthropic reports Claude Opus 4.7 achieves 46.9% on Humanity’s Last Exam (HLE) without tools. The company states this represents the highest score among current frontier models on this benchmark variant. That claim requires precision: HLE results differ significantly depending on whether models use tool access. MiniMax M2.7, for instance, is reported to score 44.9% on HLE with tools, a different variant, not directly comparable to Opus 4.7’s without-tools figure. Readers evaluating benchmark claims should note the variant distinction before drawing competitive conclusions. Independent verification of the 46.9% figure through Epoch AI’s evaluation is pending URL confirmation; Anthropic’s report is the current source.
The more distinctive disclosure is what Anthropic said about the model’s ceiling. The company confirmed that Claude Opus 4.7 is not its most capable system. An internal model called Claude Mythos Preview exceeds Opus 4.7 in overall capability, and Anthropic explicitly stated that Opus 4.7’s cyber capabilities are less advanced than those of the internal model. This kind of disclosure – a frontier lab publicly acknowledging a gap between its safest-to-release model and its most capable internal system, reflects a deliberate release architecture, not a communications slip.
Why does this matter for developers and enterprise teams? Three things.
First, the 1M token context window is confirmed and substantial. For teams working with large codebases, extended legal documents, or complex research corpora, this is a practical capability threshold that changes what’s feasible in a single API call.
Second, the GitHub Copilot integration means Opus 4.7 is immediately accessible to a wide enterprise developer base without requiring separate API contracts. Teams already on Copilot Business or Enterprise can evaluate it today.
Third, the Claude Mythos Preview disclosure is a signal worth tracking. Anthropic’s framing suggests the company is managing a capability-safety tradeoff in its release decisions, releasing what it judges safe to release broadly, while retaining more powerful systems internally. That’s a governance posture, not just a product one. It has implications for how practitioners should think about what any given frontier lab’s “most capable public model” actually represents relative to internal capability.
The April 16 announcement previewed the release; the April 22 GA date is the operational event. Teams evaluating Opus 4.7 integration should start with the platform.claude.com release notes for confirmed capability details and review Anthropic’s announcement for positioning context.
Prior TJS coverage on Anthropic’s infrastructure position, including the AWS compute partnership – provides relevant context for understanding where this model sits in the broader Anthropic-AWS relationship. See the Markets pillar for that coverage.