Claude models lead general-purpose LLM rankings

Claude Opus 5.5 leads a field of 37 evaluated large language models with a 4.7/5 rating, followed by Claude Sonnet 5.5 and GPT-6.1 Sol at 4.6/5. Opus targets demanding coding agents, audits and document-heavy research. Sonnet offers near-Opus performance at half the token price, while Sol provides OpenAI customers with near-Astra results at a fifth of the token price, but slower output.

Lower-cost models offer alternatives for high-volume workloads. Claude Haiku 5.5 scores above Gemini 3.8 Flash on independent tests at a sixth of the cost per task. GPT-6 Luna targets classification, extraction and chat within OpenAI-based applications. Among open-weight options, Qwen3.8-27B combines an Apache licence with coding capabilities and fits on a single 24 GB card at 4-bit. GLM-5.3-Flash offers an MIT-licensed option for coding and multimodal agents.

Deployment constraints remain important alongside benchmark performance. DeepSeek requires accepting China-based API handling or arranging alternative hosting. Some models rely on vendor-published results without independent evaluation, while others impose commercial licensing restrictions. Google’s Gemini models retain appeal for audio, video and document processing, but face price and response-speed disadvantages. Model selection depends on workload, infrastructure, data handling and the ability to verify outputs.