ResearchBenchmarksAutoregressive Generation on Long-context evaluation document 122,466 tokensAutoregressive Generation on Long-context evaluation document 122,466 tokensFollow29.3Time To First Token (ms)Nectar8.192150.671293.15435.629May 10, 2026Evaluation resultsMethodMethodLinksTime To First Token (ms)Throughput (tokens/s)Memory Usage (GB)NectarModel=Qwen3-1.7B, Infe...Model=Qwen3-1.7B, Inference Protocol=Nectar2026.0529.337.78.2NectarModel=Qwen3-4B, Infere...Model=Qwen3-4B, Inference Protocol=Nectar2026.0550.115.717.1BaseModel=Qwen3-1.7B, Infe...Model=Qwen3-1.7B, Inference Protocol=Base2026.052994.460.4BaseModel=Qwen3-4B, Infere...Model=Qwen3-4B, Inference Protocol=Base2026.055571.584.2