What Reflection AI does
Reflection AI is an open frontier AI lab building superintelligent autonomous coding agents and the open-weight models that power them. The company was founded in March 2024 by Misha Laskin, who led reward modeling for Google DeepMind's Gemini, and Ioannis Antonoglou, co-creator of AlphaGo. Its first product, Asimov, is a multi-agent code research and reasoning system that retrieves context from massive codebases using small long-context retriever agents and a large short-context combiner agent that synthesizes answers. Asimov tracks every step of software development — from spec to PR review — so engineering teams can move faster on large, unfamiliar codebases.
Who it's for
Reflection's customers are engineering organizations operating large monorepos and platform teams that need an agent capable of reasoning across millions of lines of code. The lab also positions itself as the Western, open alternative to closed-source frontier labs like OpenAI and Anthropic and to Chinese open-weight efforts like DeepSeek, so its weights and research are aimed at developers who want sovereignty over their model stack.
Pricing
Asimov is sold via enterprise contracts to engineering teams and offers a developer trial. Open-weight model releases are free to download and run, with a hosted inference offering for production workloads.
Team & funding
The team is a 79-person concentration of researchers poached from DeepMind, OpenAI, and other frontier labs. Reflection raised $130 million across earlier rounds, then closed a $2 billion round in October 2025 led by Nvidia at an $8B valuation — a 15x jump from a $545M valuation seven months prior. Investors include Disruptive, DST, 1789, B Capital, Lightspeed, GIC, Eric Schmidt, Sequoia, CRV, and Citi. Reports in early 2026 indicate the company is engaging investors at a valuation north of $20B, supported by roughly $20M in revenue from Asimov.