Anthropic released Claude Opus 4.7, its new top-of-the-line model, and the biggest story is agentic coding. Internal and partner benchmarks show Opus 4.7 making a significant leap over Opus 4.6, particularly on long-horizon multi-step coding tasks where earlier models tended to drift off-plan or abandon tool calls mid-way.
On the capability side, the release hits four pillars:
Agentic coding: markedly higher completion rates on multi-file, multi-turn engineering tasks. Opus 4.7 is more willing to plan, run, observe, and replan rather than guessing.
Instruction following: the model is better at respecting nuanced constraints — tone, scope limits, forbidden actions — without needing repeated reminders.
Multimodal: stronger native support for images and richer visual reasoning, with better OCR and diagram understanding.
Memory: improved long-term memory behavior across sessions, with more reliable recall of user preferences, project state, and prior decisions.
Opus 4.7 is rolling out through Claude.ai, the Anthropic API, and the major cloud partners. Anthropic is positioning it as the default model for serious coding agents, enterprise automation, and any workflow where a single model needs to hold a lot of context and run a long chain of tool calls without losing the thread.
Pricing and exact benchmark numbers track closely with Opus 4.6, but early users report that the real difference shows up not in one-shot prompts but in long agent runs — the tasks where previous models quietly gave up.