Anthropic unveils Claude Opus 5: a new benchmark in coding and knowledge work at half price

Anthropic officially announces new language model Claude Opus 5. According to the developers, the model demonstrates the highest level of performance, overtaking existing solutions in programming tasks and complex intellectual work.

With the same efficiency as the Fable 5, the novelty from Anthropic costs half as much, making it one of the most profitable solutions on the market.

A new level of autonomy and work with code

In programming tests, the Claude Opus 5 showed results unattainable for other models with a similar configuration. Representatives of Anthropic shared real cases of using the neural network:

1. Creating 3D models from pixels (computer vision)

As part of the experiment, the models were given a drawing of the machine part with the task of writing code for its recovery in the form of a 3D model in FreeCAD. Opus 5 did not have direct access to the drawing. To solve the problem, the model independently developed a computer vision conveyor, extracted geometry directly from the original pixels and fully restored the part. The Opus 5 has successfully repeated this process several times, while none of the competing models has succeeded even after five attempts.

2. Finding and eliminating deep bugs

Opus 5 was involved in the analysis of the code of the popular messenger. The model not only discovered the error and revealed its root cause, but also fixed a specific extreme case that was missed by community developers when creating an official patch. A competing model in the same test eliminated only superficial symptoms by erroneously reporting a complete correction.

3. Development of financial infrastructure and auto testing

An engineer at a trading company used Opus 5 to write a stream of market data for a new exchange. The task was completed in one session. No model has been able to solve this problem, even with detailed instructions and plans from an engineer. With no access to a real data stream to verify the code, Opus 5 independently created a test environment and made sure that the data processing was correct.

Outcome

Claude Opus 5 sets a new bar for ?? development assistants. The model demonstrates the ability not only to write code according to instructions, but also to create auxiliary tools (conveyors of vision, test environments) to test their own decisions.