- Sources: primary, discussion
- Summary: Every figure in the post is attributed to Artificial Analysis, a named third-party evaluator running a published composite of nine evaluations, rather than to an internal benchmark. The leading claim is scoped to European models: 43 against Mistral Medium 3.5 at 30, NVIDIA Nemotron 3 Ultra at 38 and Inkling at 42, in a field the vendor states is led by Claude Opus 5 at 63. The company also reports 75.0 on AA-LCR, 69.3 on Terminal-Bench v2.1, and 15.3 seconds to return 500 tokens including reasoning time, and states this is its first large model, running in English and Spanish through the CompactifAI API.
- Why it matters: The vendor's own headline claim is bounded to a regional field and rests on a third-party evaluator's published composite rather than on numbers it produced itself.
- Follow-up: Track whether Quasar 438B appears in evaluations outside Artificial Analysis.
send feedback on this story