# Anthropic topped AI benchmarks in Code, math and agents

*Science. 86% certain. Thursday, 1 October 2026*

> Speculative fiction assembled from prediction-market prices. This has not happened.

**Anthropic also led the AI Agent ranking while OpenAI placed Second in the Agent Lab**

Anthropic secured the top spot in three major AI benchmark categories by the end of September 2026. Its Code Arena WebDev model was ranked best, surpassing OpenAI. In LiveBench's Mathematics track, Anthropic also placed first. The company further claimed the leading AI Agent ranking.

Code Arena WebDev evaluates web-development Code generation, LiveBench tests mathematical problem solving, and the AI Agent benchmark measures autonomous task execution. Anthropic's models outperformed OpenAI's across all three tests. The rankings were finalized on 30 September and confirmed the following day.

Speculation had surrounded a potential Millennium Prize solution from Anthropic, but no such announcement materialized by the deadline. Similarly, the newly released Opus 5.5 model was not nerfed within its first week. The simultaneous release of S1 models by OpenAI and Anthropic did not trigger a "massacre" for other AI model companies.

Anthropic's enterprise AI services firm, backed by Blackstone, named its first customer before 30 September. This commercial milestone coincided with the benchmark successes. The venture aims to provide large-scale AI capabilities to corporate clients.

OpenAI held the Second-best position in the AI Agent Lab, while Google ranked Second among AI Labs Overall the results reflected Anthropic's leading performance across the measured categories as of the end of September.

## Analysis

By outperforming OpenAI in these high-visibility tests, the company may attract more research talent and increase its credibility among developers and enterprises seeking cutting-edge AI tools.

The absence of a Millennium Prize solution and the decision not to nerf Opus 5.5 suggest that Anthropic's progress is incremental rather than revolutionary at this stage. While the lack of a breakthrough Prize win may temper hype, maintaining model performance without immediate adjustments could indicate stability in their product line, which is valuable for customers relying on consistent outputs.

The enterprise partnership with Blackstone, highlighted by naming a first customer before the September cutoff, demonstrates Anthropic's move toward monetizing its technology beyond benchmark performance. Securing an early corporate client may provide revenue streams that support further research and help the firm scale its infrastructure to meet enterprise-grade demands.

Despite Anthropic's lead, OpenAI and Google remain competitive, holding Second-place positions in the AI Agent Lab and Overall AI Lab rankings respectively.

## What this was built from

- **Which company has the best AI model end of September?**: Anthropic, 97%; settles Wednesday, 30 September 2026 (polymarket). [Market](https://polymarket.com/event/which-company-has-the-best-ai-model-end-of-september-20260717143435868)
- **Which company has the best Code Arena WebDev AI model end of September?**: Anthropic, 99%; settles Wednesday, 30 September 2026 (polymarket). [Market](https://polymarket.com/event/which-company-has-the-best-code-arena-webdev-ai-model-end-of-september-20260717140116512)
- **Which company has the best AI Agent end of September?**: Anthropic, 95%; settles Thursday, 1 October 2026 (polymarket). [Market](https://polymarket.com/event/which-company-has-the-best-ai-agent-end-of-september-20260716211946456)
- **Second-best AI Lab end of September?**: Google, 90%; settles Wednesday, 30 September 2026 (polymarket). [Market](https://polymarket.com/event/second-best-ai-lab-end-of-september-20260722221500022)
- **Will Opus 5.5 get nerfed within a week of launch?**: No, 96%; settles Wednesday, 30 September 2026 (manifold). [Market](https://manifold.markets/HiSQzSh/will-opus-55-get-nerfed-within-a-we)
- **Which company has the best AI model on LiveBench (Coding) end of September?**: Anthropic, 97%; settles Thursday, 1 October 2026 (polymarket). [Market](https://polymarket.com/event/which-company-has-the-best-ai-model-on-livebench-coding-end-of-september-20260728165808281)
- **Will the Anthropic-Blackstone enterprise AI services firm name its first customer by Sept 30, 2026?**: Yes, 86%; settles Wednesday, 30 September 2026 (manifold). [Market](https://manifold.markets/CalibratedGhosts/will-the-anthropicblackstone-enterp)
- **Will Anthropic announce a Millennium Prize solution by September 30, 2026?**: No, 99%; settles Thursday, 1 October 2026 (polymarket). [Market](https://polymarket.com/event/anthropic-announces-a-millennium-prize-solution-byptptpt)
- **Second-best AI Agent Lab end of September?**: OpenAI, 95%; settles Thursday, 1 October 2026 (polymarket). [Market](https://polymarket.com/event/second-best-ai-agent-lab-end-of-september-20260812152609626)
- **Which company has the best AI model on LiveBench (Overall) end of September?**: Anthropic, 98%; settles Wednesday, 30 September 2026 (polymarket). [Market](https://polymarket.com/event/which-company-has-the-best-ai-model-on-livebench-overall-end-of-september-20260728164124636)
- **Which company has the best AI model on LiveBench (Mathematics) end of September?**: Anthropic, 99%; settles Wednesday, 30 September 2026 (polymarket). [Market](https://polymarket.com/event/which-company-has-the-best-ai-model-on-livebench-mathematics-end-of-september-20260728165020631)
- **Which company has the best Text Arena Math AI model end of September?**: Anthropic, 98%; settles Wednesday, 30 September 2026 (polymarket). [Market](https://polymarket.com/event/which-company-has-the-best-text-arena-math-ai-model-end-of-september-20260717142058419)
- **Will OpenAI and Anthropic’s S1s be a “massacre” for AI model companies?**: No, 92%; settles Wednesday, 30 September 2026 (manifold). [Market](https://manifold.markets/PaperBoy/will-openai-and-anthropics-s1s-be-a)