Anthropic releases Claude Opus 4.7, claiming top results in coding and agentic tasks

Anthropic has released Claude Opus 4.7, betting that the next phase of competition between AI models will be decided not only by answer quality but by the ability to run long work chains reliably. The new version is positioned as the company's most capable generally available model: it writes and fixes code better, holds multistep context more firmly and makes noticeably fewer mistakes when working with tools. It matters for the market also because this is not an experimental showcase but a model that can be bought and built into workflows right now.

The release's main argument is its result on SWE-bench Pro, one of the most prominent benchmarks for judging whether models can solve real engineering tasks. According to the company, Claude Opus 4.7 scored 64.3%, while GPT-5.4 scored 57.7%. It is an important signal: the focus is no longer on a model's abstract "smartness" but on how well it can understand a codebase, find bugs, propose patches and carry a task through to a working result. Teams that bring AI into development, support and internal automation watch such tests especially closely.

Anthropic's second emphasis is agentic behavior. The company describes stronger coordination of multiple agents in scenarios that can run for hours. These are tasks in which the model does not simply answer a single request but plans steps, calls tools, checks intermediate results and keeps working without constant human intervention. This is where the gap between a flashy demo and a system that can be built into a real process shows up most often: the longer the chain of actions, the costlier errors, lost context and wrong tool calls become.

Compared with previous versions, Anthropic also claims a 14% improvement in multistep agentic reasoning and three times fewer tool errors. If these figures hold up in practice, they may matter even more than the gap on any single benchmark. For enterprise users, reliability is usually worth more than peak quality: if a model "breaks" the workflow a little less often, does not lose state and calls external services more correctly, it is easier to trust with operations involving code, analytics, documents and internal bots.

The company also points to a threefold increase in image resolution, which widens the range of uses for reading schematics, interfaces, diagrams and other visually dense material.

Pricing stays within the range Anthropic already uses for its top-tier models: $5 per million input tokens and $25 per million output tokens. This does not look like an attempt to undercut the market, but it makes the release easy to understand for existing customers: the company is selling not just another quality bump but a more reliable tool for complex work. For teams that count costs per completed task rather than only per token, this may be a strong argument: a more stable model requires fewer manual checks, repeat runs and fixes after failed calls.

Against the backdrop of the race between Anthropic, OpenAI, Google and other players, the move looks logical. The winner now is not whoever talks loudest about "general intelligence" but whoever's model handles applied tasks best: writing code, driving tools, sustaining long sessions and delivering predictable results in production conditions.

The takeaway is simple: Claude Opus 4.7 is not a cosmetic update but Anthropic's bid for leadership in models for development and agentic automation. If the promised improvements match what teams see in production, pressure on competitors will grow, not because of polished comparison tables but because of a more practical question: which model is cheaper and safer to put at the center of a real workflow.