Reflection's Beam, billed as a Western open-weight frontier, trails Qwen and DeepSeek on coding tests
TheTuringPost · x · 2026-10-06
Reflection launched its new model Beam, claiming it "advances the Western open-weight frontier," but TheTuringPost found it well behind leading Chinese open models on key coding benchmarks — a far cry from co-founder Ioannis Antonoglou's stated February goal of building "the best open models out there."
More controversially, Reflection's own comparison table omitted six Qwen and DeepSeek results that the author verified from other sources; adding them back makes the gap even starker, undercutting the "Western frontier" framing.
Related event: US Open-Source Model Beam Falls Short of DeepSeek and Qwen in Benchmarks(3 posts)→
More from Companies & People
- Drone flying and coding now mandatory in 1,200+ Chinese vocational schools — alexcovo_eth · 2026-10-06
- Sam Altman shares Michael Moritz's five principles behind Sequoia's decades of dominance — darian314 · 2026-10-06
- Lee Kai-fu brokers Kazakhstan's school AI pilot: 10 schools now, 500 by April 2027 — kaifulee · 2026-10-06
- n8n Grew 10x Last Year Despite Being Declared Dead Three Times, CEO Says — aakashgupta · 2026-10-06
- AI companies often voice support for state AI bills only after they pass, observer notes — Miles_Brundage · 2026-10-06
- Skip SF Tech Week events and host your own, says founder who turned one birthday party into 10+ real connections — tech__unicorn · 2026-10-06