Digital Economy

Anthropic releases Claude Opus 4.1 while the market awaits GPT-5

The start-up updates its flagship model at a key moment for the generative AI sector

7 August 2025

2' min read

Anthropic has released Claude Opus 4.1, an incremental update of its flagship language model that comes at a peculiar time for the market: while the industry still awaits concrete news about OpenAI's GPT-5, scheduled according to some sources for the end of August 2025.

The update, described by the company as a 'drop-in replacement' for Opus 4, brings improvements in coding and agent task performance, reaching 74.5 per cent accuracy in software engineering benchmarks, compared to 72.5 per cent in the previous version.

An evolving market

The release is part of an increasingly competitive environment. OpenAI launched GPT-4.1 in April with a score of 54.6 per cent on the same SWE-bench benchmark, while work continues on GPT-5. Google regularly updates its Gemini line, and Meta recently released the Llama 4 models with multimodal capabilities.

Each market player is looking for its own niche: OpenAI focuses on integration and general reasoning, Google excels in multimodal understanding, while Anthropic is positioning itself on professional coding applications and sustained performance over time.

The price of Claude Opus 4.1 remains unchanged from the previous version: $15 per million tokens in input and $75 in output. This premium positioning reflects the product's enterprise target, considering that OpenAI's GPT-4.1 costs $2 in input and $8 in output.

Technical improvements

According to Anthropic's internal tests, the new model shows particularly good progress in refactoring multi-file code and debugging complex codebases. Rakuten tested the system on a seven-hour open source refactoring project, reporting stable performance throughout the test.

GitHub announced the integration of Claude Opus 4.1 into its Copilot for Enterprise and Pro+ users, offering an alternative to the OpenAI models already available. A move that reflects the platforms' strategy to diversify its AI model offering.

The improvements were described as incremental but significant: Windsurf, a platform for developers, measured a performance gain of one standard deviation compared to Opus 4, comparable according to them to the jump between previous generations of models.

The wait for GPT-5

Meanwhile, the industry continues to wait for news on GPT-5. Sam Altman, CEO of OpenAI, initially mentioned summer 2025 as the release date. According to Axios, sources close to the company are now talking about August, although the exact date remains uncertain.

OpenAI has not stood still: in February, it released GPT-4.5 as a 'research preview', followed by the o3 and o4-mini series models. However, the company itself made it clear that these do not represent the generational leap expected with GPT-5.

Expectations on GPT-5 remain high. According to rumours, it should integrate advanced reasoning capabilities, full multimodal support and a significantly expanded context window. But until the actual release, only speculation remains.

A look into the future

Claude Opus 4.1 is available through the Anthropic API, Amazon Bedrock and Google Cloud Vertex AI. This cross-platform availability facilitates integration for companies already using these cloud services. Anthropic said it plans 'substantial improvements' in the coming weeks, suggesting that Opus 4.1 could be part of a broader roadmap of updates.