Bobo Li

Research Fellow @ National University of Singapore | LLMs | NLP | AI Agents | Cognitive AI

[Bobo Li @Bobo_hashtag]

@NFT_Chen 可以看看我们的 OneJev，27B速度只需 32ms，0.8b 速度 5ms
https://t.co/ITTn6C2yOT

[Bobo Li @Bobo_hashtag]

We just released OneJev 🔥, an open multimodal System One decision model.
🌐 https://t.co/vuGgdinslu

1. OneJev-27B matches the official Jev on TypeSafe (89.3 vs 89.0), while OneJev-0.8B already outperforms Jev-Omni 12B (70.0 vs 60.8).
💻 https://t.co/0r5fUxjpNE

2. OneJev works with screenshots, photos, video, and text, and comes in four sizes: 0.8B / 4B / 9B / 27B.

3. OneJev is built for fast, repeated decisions. On one H200, OneJev-0.8B takes just 31 ms for one question. OneJev-4B takes 64 ms for one question and 104 ms for ten questions.

4. More details
🤗 https://t.co/KdLk9LgNQS
💻 https://t.co/0r5fUxjpNE
🌐 https://t.co/vuGgdinslu

Built on Qwen3.5 and Qwen3.8. Thanks @Alibaba_Qwen for the open base models 🙏

[Bobo Li @Bobo_hashtag]

@NFT_Chen 也欢迎关注我们的 OneJev ，支持多模态和本地部署， 0.8B 响应速度只需要 30ms，同尺寸模型中最快， 27B 只需 189ms，TypeSafe 上效果超过 Jev1.13

https://t.co/Lx4HGKlrop

[Bobo Li @Bobo_hashtag]

Full results across models and benchmarks.

[Bobo Li @Bobo_hashtag]

We just released OneJev 🔥, an open multimodal System One decision model.
🌐 https://t.co/vuGgdinslu

1. OneJev-27B matches the official Jev on TypeSafe (89.3 vs 89.0), while OneJev-0.8B already outperforms Jev-Omni 12B (70.0 vs 60.8).
💻 https://t.co/0r5fUxjpNE

2. OneJev works with screenshots, photos, video, and text, and comes in four sizes: 0.8B / 4B / 9B / 27B.

3. OneJev is built for fast, repeated decisions. On one H200, OneJev-0.8B takes just 31 ms for one question. OneJev-4B takes 64 ms for one question and 104 ms for ten questions.

4. More details
🤗 https://t.co/KdLk9LgNQS
💻 https://t.co/0r5fUxjpNE
🌐 https://t.co/vuGgdinslu

Built on Qwen3.5 and Qwen3.8. Thanks @Alibaba_Qwen for the open base models 🙏

[Bobo Li @Bobo_hashtag]

Fast decisions in milliseconds.

[Bobo Li @Bobo_hashtag]

OneJev in one screen.

[Bobo Li @Bobo_hashtag]

Strong performance across benchmarks.

[Bobo Li @Bobo_hashtag]

Try OneJev here 👇

💻 GitHub: https://t.co/Lx4HGKlrop

[Bobo Li @Bobo_hashtag]

@Apodex_AI While Apodex is fantastic, OmniScientist is also worth a try: https://t.co/ZEhKubMZDg

[Bobo Li @Bobo_hashtag]

5/5 OmniScientist is open-source and comes in three forms.
🖥️ Desktop for interactive research
⌨️ CLI for local workflows
🧩 Skill for agent workflows
🌐 https://t.co/ZEhKubMZDg
💻 https://t.co/CzyI5QnEkY

[Bobo Li @Bobo_hashtag]

Meet OmniScientist 🧑‍🔬 An open-source AI scientist for end-to-end research on raw multimodal data.
Raw evidence → hypotheses → experiments → results → paper.
Now available in three forms
🖥️ Desktop
⌨️ CLI
🧩 Skill

📄 https://t.co/0GeEDnHWvl

[Bobo Li @Bobo_hashtag]

4/5 We also built traceability into the research process.
Numerical claims can be traced back to actual execution records. Analyses remain inspectable from experiment to manuscript. ✅
This makes the generated research much easier to verify and reproduce.

[Bobo Li @Bobo_hashtag]

3/5

We evaluated OmniScientist on 36 real-data research cases across 5 discipline families.

It completed 36/36 cases end to end.

Direct perception also improved all 7 evaluation dimensions compared with using precomputed features. 👁️

[Bobo Li @Bobo_hashtag]

2/5

OmniScientist covers the full research loop.
Raw data → evidence inspection → hypothesis → analysis code → execution → result interpretation → manuscript
The system can keep iterating as new results come in. 🔁

[Bobo Li @Bobo_hashtag]

1/5 OmniScientist works directly with raw scientific evidence 🔬

Images · signals · audio · video
3D structures · trajectories
Tables · formulae · graphs

[Bobo Li @Bobo_hashtag]

Source Code👉 https://t.co/CzyI5QnEkY

[Bobo Li @Bobo_hashtag]

Meet OmniScientist 🧑‍🔬 Your open-source AI scientist workbench for end-to-end research.
Raw multimodal data → hypotheses → experiments → results → paper. All in one place.
🖥️ Desktop · ⌨️ CLI · 🧩 Skill
📄 https://t.co/0GeEDnHWvl
🌐 https://t.co/ZEhKubMZDg