Reflection AI

Reflection AI released an open-source model called Beam, with approximately 501 billion total parameters and only 23 billion activated parameters, using a MoE architecture, designed specifically for reasoning, coding, and agentic workflows, trained end-to-end from scratch, with weights to be made available this month.

According to Axios, startup Reflection AI is about to release a domestically developed American open-weight large model.

Beam currently cannot be run by users not on Reflection's early access waitlist; the model is undergoing final red-teaming, and the full weights, technical report, and model card are expected to be released later in October 2026, with pricing and API terms not yet announced.

Reflection AI self-reports that Beam scores 80.9 on SWE-Bench Verified, 80.1 on Terminal-Bench 2.1, and 97.8 on AIME 2026.

The above benchmark results for Beam are self-reported data, had not been independently verified at the time of publication, and a complete side-by-side scorecard with DeepSeek, Qwen, Kimi, etc. had not been published at the time of release.

Reflection AI is a Brooklyn-based startup founded by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou.

Kepler's core team members come from companies such as Palantir, Meta, Goldman, and Reflection AI, among them Sarah, who was one of the engineering leads for Palantir Ontology, and Chris, who worked at Palantir for 15 years focusing on optimization problems.

Reflection claims Beam's reasoning ability is comparable to GLM-5.2, but its reasoning compute is about 1/3 to 1/4 of the latter's.

Reflection AI was founded by two former Google DeepMind researchers, its models focus on coding and agent directions, and are optimized for Nvidia chips.

At present this deal is still in the early negotiation stage, there is a possibility that the negotiations break down, and the target's valuation is high, so once it advances it will face strict regulatory review.

Reflection AI was founded by former DeepMind researchers.

Reflection AI, as a newer participant, chose the open-source route rather than closed-source.

The acquisition is still in the negotiation stage and has not been finalized, with key data such as deal value and team size unavailable.

Reflection AI claims Beam is on par with the GLM series while requiring 3 to 4 times less inference compute; the company has raised approximately $4.7 billion in cumulative investment from NVIDIA, Sequoia, and others, though the performance figures are self-reported and the weights have not been released.

Reflection AI was founded in late 2024 and focuses on a safety-first model development approach.

Reflection AI completed a $2 billion funding round, with investors including NVIDIA, Sequoia, Lightspeed, DST, Citi, and Eric Schmidt.

Reflection officially acknowledged that frontier open-source models such as Kimi K3 still lead Beam in raw capability, and that Beam's advantage lies in efficiency.

Reflection stated that Beam achieves comparable scores to GLM on reasoning benchmarks while consuming three to four times less inference compute, a figure estimated based on activated parameters and generated token counts rather than measured costs.

Reflection AI raised billions of dollars and signed a massive compute agreement with SpaceX, along with a contract worth approximately one billion dollars with Nebius.

Reflection AI's new model targets DeepSeek and Alibaba's Qwen series, the leaders in the open-source domain.