Reflection Beam: Open-Weight AI Model Takes On Chinese Rivals

Reflection says Beam scores on par with Z.ai's GLM-5.2 on advanced reasoning benchmarks and outperforms today's leading Western open models. Chief executive Misha Laskin told Semafor that Beam is "approaching" the performance of Alibaba's Qwen 3.8-Max, which itself trails the top closed American models from OpenAI and Anthropic.

The most direct US rival may be Inkling, the open model from Mira Murati's Thinking Machines Lab. Reflection's own benchmarks show Beam scoring higher on four coding tests where both report results, though Inkling accepts multiple input types while Beam handles text only. The performance claims have not been independently verified.

Reflection is also working with the US Center for Advancing Innovation and Standards for Super Intelligence and the UK's AI Safety Institute, according to Semafor.

Reflection raised roughly $4.7 billion from backers including Nvidia, Sequoia Capital and Lightspeed Venture Partners, and confirmed in June a $25 billion pre-money valuation. The company has also secured compute: this summer it signed deals worth more than $7 billion with SpaceX and Nebius for access to Nvidia's GB300 chips through 2029.

Its commercial pitch centres on "AI factories" — local systems that institutions can train on their own private data. Reflection has begun testing the concept with South Korea's Shinsegae Group, and Axios reported that hedge funds and trading firms are among those interested. The startup says Beam's weights and full technical details will be released this month, with distribution through hyperscalers and cloud providers.

Laskin said Reflection is already training its next model, which he described as "much more" powerful than Beam.

For now, Beam is aimed at companies and public institutions, not consumers, and Reflection has not announced any availability or pricing in Tunisia. The open-weight release is the part that matters locally: once the weights and technical details are published this month, developers and researchers will be able to download, inspect and run the model on their own infrastructure, subject to the licence Reflection attaches.

If the claimed efficiency holds up under outside testing, models like Beam could lower the cost of building Arabic-language and Tunisian-focused AI tools, since running a model needs less hardware than comparable open alternatives. Treat the benchmark numbers as company claims until independent evaluations appear.