VQV Signal
According to a post on X by @baseten, Baseten and Genspark post-trained an open-weight model starting from Qwen3.5-397B. They say it scores 0.759 on Genspark's internal deep-research evals, ahead of Opus 5.5 at 0.738 and the base model at 0.612. The research stage was trained with a reward that credits breadth of searches and sources, while report writing was handled separately.
According to a post on X by @baseten, Baseten and Genspark post-trained an open-weight model starting from Qwen3.5-397B. They say it scores 0.759 on Genspark's internal deep-research evals, ahead of Opus 5.5 at 0.738 and the base model at 0.612. The research stage was trained with a reward that credits breadth of searches and sources, while report writing was handled separately.
AI-assisted summary based on the listed source.
Starting from Qwen3.5-397B, Baseten and Genspark say their post-trained model scores 0.759 on Genspark's internal deep-research evals, ahead of Opus 5.5 (0.738) and the base model (0.612). They trained research with a reward that credits breadth of searches and sources, and handled report writing separately. Why...
The result suggests open-weight models can, with tuning, outscore a closed frontier model on this internal benchmark. Teams building research agents may want to watch whether the gains hold on public evaluations and independent tests. The figures come from the companies' own internal evals and have not been independently confirmed here.
VQV organizes public signals from inspectable sources. It does not independently verify the underlying report.
Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.
VQV surfaced this signal because it is recent, relevant to Open Source LLMs, connected to Scobleizer News Pulse (X).
No login, cookies, social SDKs, or automatic posting.
