Science
Anthropic topped AI benchmarks and landed first enterprise client
The firm also achieved a 1520 Overall arena score, outpacing rivals such as OpenAI and Alibaba.
By the end of September 2026 Anthropic had emerged as the leading AI model developer. Its models outperformed rivals across several benchmark arenas, including OpenAI and major Chinese firms. The company also announced its first enterprise client through its joint venture with Blackstone.
Anthropic's model achieved an Overall arena score of 1520, the highest recorded for the period. In the LiveBench coding competition it held the top position. With Style-Control enabled it was ranked number one across all categories. These results placed the firm ahead of competitors such as Meta and DeepSeek.
The firm's AI Agent was judged the best in its category at the end of September. Its text-based math model also claimed the top spot in the text Arena. In the Code Arena WebDev track Anthropic secured the Second-best position, trailing only the leading Lab OpenAI occupied lower rankings in these assessments.
Anthropic's partnership with Blackstone named its inaugural enterprise customer before September 30.
Among Chinese companies, Alibaba was identified as the top performer in the September rankings. Despite Alibaba's lead in the domestic sector, Anthropic's scores remained superior on the global stage. The gap reinforced Anthropic's dominance over both Western and Chinese rivals.
Analysis
By posting the highest Overall arena score and leading in coding, text-math, and Agent categories, the firm has demonstrated technical superiority that could attract further research collaborations and talent.
The 1520 Overall score provides a quantifiable measure of performance that may influence downstream adoption decisions by developers and enterprises. Leading positions on LiveBench and in Style-Control-enabled tests suggest robustness across both Code generation and creative tasks, potentially widening Anthropic's appeal beyond niche applications.
What this was built from
Each claim above traces to a market below, at the price it was trading when this edition went to press.