Genspark deploys post-trained Qwen3.5-397B for Deep Research
Baseten reports that Genspark worked with Baseten Applied Research to post-train Qwen3.5-397B into a specialized Deep Research model now powering Genspark’s production agent. On Genspark’s internal quality grader, the model scored 0.759, above Opus 5.5 at 0.738 and the Qwen3.5-397B base model at 0.612; Baseten also says it researches across five times as many sources and runs with NVFP4 weights on two NVIDIA B200 GPUs per replica. These benchmark and deployment figures come from the vendors and were not independently reproduced in the supplied report.
The collaboration demonstrates how an open-weight 397B model can be specialized for a production research agent, but the claim that it surpasses closed frontier models rests on Genspark’s own grader.
News
How Baseten and Genspark trained a deep research model to outperform Opus 5.5
Genspark worked with Baseten Applied Research to post-train Qwen3.5-397B into a specialized deep research model that outperforms the closed frontier.
Authors
Date
November 10, 2026
#Genspark#Baseten#Qwen3.5-397B#Baseten Applied Research#Baseten Loops SDK#Opus 5.5#Sonnet 5.5#GPT-5.6 Sol#Kimi K3#DeepSeek V4 Pro#NVIDIA B200