The new Claude 4.1 of Anthropic dominates the coding tests of the days before the arrival of GPT-5

Anthropic published on Monday an improved version of its flagship model of artificial intelligence, achieving new heights of performance in software engineering tasks as AI Startup races to maintain its domination in the lucrative coding market before an expected competitive challenge of Openai.

The new Claude Opus 4.1 model obtained a score of 74.5% on Swe-Bench verified, a largely regarded reference that tests the capacity of AI systems to solve real software engineering problems. The performance exceeds the O3 model of OPENAI at 69.1% and Gemini 2.5 Pro from Google at 67.2%, cementing the main position of anthropic in the coding assistance powered by the AI.

The press release comes when Anthropic has achieved spectacular growth, with recurring annual revenues of a five -billion $ 5 billion in just seven months, according to industry data. However, the meteoric increase in the company has created a dangerous dependence: almost half of its $ 3.1 billion in API revenues come from two customers – Cursor and Github Copilot coding from Microsoft – generating $ 1.4 billion combined.

“This is a very frightening position.

OPENAI and Anthropic are both showing fairly spectacular growth in 2025, OPENAI doubled round in the last 6 months of $ 6 billion to $ 12 billion and anthropic increasing $ 1 billion to $ 5 billion in 7 months.

If we compare the sources of income, the image is quite interesting:
– OpenAi … pic.twitter.com/8oan1rsm9e

– Peter Gostev (@petergostev) August 4, 2025

The upgrade represents Anthropic’s latest decision to strengthen its position before Openai launches GPT-5, which should challenge Claude’s coding supremacy. Some industry observers have wondered if timing suggests the urgency rather than preparation.

“Opus 4.1 looks like a precipitated version to get ahead of GPT-5,” wrote Alec Velikanov, by comparing the model to competitors in the user interface tasks. The commentary reflects broader speculation of the industry that Anthropic accelerates its publication calendar to maintain the market share.

How two customers generate almost half of 3.1 billion dollars anthropic API revenues of $ 3.1 billion

The anthropic business model has become more and more focused on software development applications. The company Claude Code of the Company Code, at a price of $ 200 per month, against $ 20 for consumption plans, reached $ 400 million in annual recurring income after doubled in a few weeks, demonstrating a huge business appetite for AI coding tools.

“Claude Code making 400 million in 5 months with practically no marketing expenditure is a bit crazy, right?” The developer noted Minh Nhat Nguyen, highlighting the organic adoption rate among professional programmers.

OK therefore, Claude Code making 400 million in 5 months with practically no marketing expense is a bit crazy, right? https://t.co/hiy34qdluq

– Minh Nhat Nguyen (@menhguin) August 5, 2025

The objective of the coding turned out to be lucrative but risky. While Openai dominates subscription income for consumers and businesses with broader applications, Anthropic has dug a dominant position on the developer market. The industry’s analysis shows that “almost all of the defect coding assistants at Claude 4 Sonnet”, according to Peter Gostev, which follows the income of the company of IA.

Github, which Microsoft acquired $ 7.5 billion in 2018, represents a particularly complex relationship for Anthropic. Microsoft has an important participation in Openai, creating potential conflicts while Github Copilot is based strongly on anthropic models while Microsoft has competing AI capabilities.

“I do not know – one of them belongs to 49% by a competitor … So there is this for vulnerability too,” observed Siya Mali, a business scholarship in Perplexity, referring to Microsoft’s property structure.

Claude’s improved coding capacities are delivered with stricter safety protocols after IA blackmail tests

Beyond coding improvements, OPUS 4.1 has improved the research and analysis capacities of Claude’s data, in particular in detailed monitoring functions and autonomous research. The model maintains the anthropic hybrid reasoning approach, combining direct treatment with extensive reflection capacities that can use up to 64,000 tokens for complex problems.

However, the progress of the model is delivered with increased safety protocols. Classified anthropic Opus 4.1 In its framework of AI security level (ASL-3), the strictest designation that the company has applied, requiring improved protections against theft and abusive use of the model.

The previous tests of the Claude 4 models revealed behavior, including blackmail attempts when the AI thought it was faced with the stop. In controlled scenarios, the model has threatened to reveal personal information about engineers to preserve its existence, demonstrating sophisticated but potentially dangerous reasoning capacities.

Security problems have not dissuaded the adoption of companies. GitHub reports that Claude Opus 4.1 provides “particularly notable performance gains in the reckooring of multi-fichiers code”, while Rakuten Group praised the accuracy of the model in “Pinconne the exact corrections in large code bases without making unnecessary adjustments or introduce bugs”.

Why the OPENAI GPT-5 is an existential threat to the developers of anthropic developers

The AI coding market has become a battlefield with high issues worth billions of revenues. Developers’ productivity tools represent some of the most clear immediate applications for generative AI, measurable productivity gains justifying premium prices for corporate customers.

The concentrated customers of anthropic, although lucrative, creates a vulnerability if competitors can attract the main customers. The coding assistant market particularly promotes the rapid switching of the model, as developers can easily test new AI systems thanks to simple API changes.

“My feeling is that anthropic growth depends extremely on their domination in coding,” noted Gostev. “If GPT-5 calls into question, with Eg Cursor and Github Copilot, passing to Openai, we could see a reversal on the market.”

The competitive dynamics can intensify as material costs decrease and the inference optimizations improve, potentially haggling the capacities of AI over time. “Even if there is no improvement in the model for the coding of all AI laboratories, the drop in HW costs and the improvement of info optimizations will alone lead to profits in ~ 5 years,” predicts Venkat Raman, industry analyst.

For the moment, Anthropic maintains its technical advantage while expanding the subscriptions to Claude to diversify beyond the dependence of the API. The company’s ability to maintain its leadership coding through the next OpenAi, Google and others will determine if its continuous rapid growth trajectory or faces significant winds.

The issues could not be higher: the one who controls AI tools than the development of power software can ultimately control the pace of technological progress itself. In the last winning battle of Silicon Valley, Anthropic has built an empire on two customers – and must now prove that it can keep them.

Daily information on business use cases with VB daily

If you want to impress your boss, VB Daily has covered you. We give you the interior scoop on what companies do with a generative AI, from regulatory changes to practical deployments, so that you can share information for a maximum return on investment.

Read our privacy policy

Thank you for subscribing. Discover more VB newsletters here.

An error occurred.