Software teams building agentic AI workflows have been pushing frontier models toward longer, unsupervised task runs. Claude Opus 4.7, now generally available from Anthropic, is aimed squarely at that demand, with particular gains in software engineering, multimodal processing, and the kind of instruction fidelity that matters when a model is running tasks autonomously over multiple steps.
Opus 4.7 is available across all Claude products and the API, Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry. Pricing remains the same as Opus 4.6: $5 per million input tokens and $25 per million output tokens.
Opus 4.7 is a notable improvement on Opus 4.6 in advanced software engineering, with particular gains on the most difficult tasks. The model handles complex, long-running tasks with rigor and consistency, pays precise attention to instructions, and devises ways to verify its own outputs before reporting back.
On the vision side, the upgrade is significant. Opus 4.7 can accept images up to 2,576 pixels on the long edge, approximately 3.75 megapixels, more than three times as many as prior Claude models.
That increase supports use cases including computer-use agents reading dense screenshots, data extractions from complex diagrams, and work that needs pixel-perfect references
The higher resolution is a model-level change, meaning images sent to the API are automatically processed at greater fidelity; users who do not need the extra detail can downsample images before sending to control token costs.
Instruction-following behavior has also shifted in ways that require attention from teams migrating existing deployments. Where previous models interpreted instructions loosely or skipped parts entirely, Opus 4.7 takes the instructions literally. Users should re-tune their prompts and harnesses accordingly.
Opus 4.7 is also better at using file system-based memory. It remembers important notes across long, multi-session work, and uses them to move on to new tasks that, as a result, need less up-front context.
The release carries specific policy weight tied to Anthropic’s earlier work on AI and cybersecurity risk. Opus 4.7 is the first model on which Anthropic is testing new cyber safeguards on a less capable model before moving toward a broader release of Mythos-class models.
Its cyber capabilities are not as advanced as those of Mythos Preview; during training, Anthropic experimented with efforts to differentially reduce these capabilities.
Opus 4.7 ships with safeguards that automatically detect and block requests indicating prohibited or high-risk cybersecurity uses. What Anthropic learns from real-world deployment of these safeguards will inform its eventual goal of a broad release of Mythos-class models.
Security professionals who wish to use Opus 4.7 for legitimate cybersecurity purposes, such as vulnerability research, penetration testing, and red-teaming, are invited to join Anthropic’s new Cyber Verification Program.
Opus 4.7 shows a similar safety profile to Opus 4.6, with low rates of concerning behavior such as deception, sycophancy, and cooperation with misuse.
On some measures, including honesty and resistance to malicious prompt injection attacks, Opus 4.7 is an improvement on Opus 4.6. In others, such as its tendency to give overly detailed harm-reduction advice on controlled substances, Opus 4.7 is modestly weaker.
Anthropic’s alignment assessment concluded that the model is “largely well-aligned and trustworthy, though not fully ideal in its behavior.” Mythos Preview remains the best-aligned model Anthropic has trained according to its evaluations. Full safety evaluations are covered in the Claude Opus 4.7 System Card.
Teams upgrading from Opus 4.6 should expect changes in token consumption. Opus 4.7 uses an updated tokenizer that improves how the model processes text, with the same input mapping to more tokens, roughly 1.0 to 1.35 times depending on content type. Opus 4.7 also thinks more at higher effort levels, particularly on later turns in agentic settings, which improves reliability on hard problems but produces more output tokens.
Anthropic says the net effect has been favorable in internal testing on a coding evaluation, with token usage improved across all effort levels.
Guide: Breach and Attack Simulation & Automated Penetration Testing
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。