Krea 2 Large is a specialized image model rather than a general-purpose reasoning or coding system. Its Artificial Analysis Image Arena Quality Elo of 1191.37 and official emphasis on aesthetics, style references, and moodboards support above-median image capability and strong creative input control. This makes its multimodal/I-O score materially stronger than Hermes 4.3 36B’s 55, though it is not comparable to Claude Opus 4.8 or Grok 3 for general reasoning, coding, or agent work. Official API pricing is unusually clear: $0.060 per text-to-image generation, $0.065 with style references, and $0.070 with moodboards. Those rates support a solid, not exceptional, cost score absent directly supplied cross-model price and throughput comparisons. API documentation and a public Krea GitHub organization indicate usable developer access, but the supplied evidence does not establish SDK maturity, regional availability, reliability commitments, or production support. Artificial Analysis covers generation time, but no concrete latency figure was supplied here. Public validation is narrow. Contra Labs found Krea 2 Large second of four on style fidelity, while its 39.5% pairwise win rate ranked fourth; this is a useful but limited preference signal, not a general quality leaderboard. DeepSWE and LiveCodeBench list no result, SWE-bench has no direct evidence, and no Terminal-Bench or Aider result was supplied. Scores therefore reflect a credible image-specialist offering with materially weaker evidence breadth than the published Claude and Grok anchors.
