Reflection
3 Intel
Reflection AI is an AI lab building open-weight frontier models, with the stated mission of making open models that anyone can access, use and build on. It was founded in March 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou, a founding engineer on AlphaGo. It initially focused on autonomous coding agents before broadening to large open models, and in 2025 raised a $2 billion round led by Nvidia at an $8 billion valuation.
Is this you? Sign in with X to claim this profile.
Intel
Reflection describes safety training for its open-weight Beam model and publishes Iris Safety Evaluations and an Open Safety Framework covering harm categories, prompt injection and capability-based risk of open weights.
Reflection released Beam, a sparse MoE with 501B total and 23B active parameters, under Apache 2.0. It ships BF16, FP8 and NVFP4 weights, a speculative drafter and the Mirror CLI harness, with day-zero serving support across major inference stacks and 40+ partner platforms.
Reflection introduces Beam, a sparse MoE with 501B total and 23B active parameters, pretrained on 23.8T tokens with large-scale RL. Benchmarks are compared against GLM 5.2, Kimi K3 and Qwen 3.8 Max. Weights and a technical report are due this month.