---
title: "SJTU's BigBang-v1 Beats a 1.6T Model Using Only Machine-Generated Training Data"
subtitle: "A 35B open model post-trained on self-generated adversarial tasks matches trillion-parameter systems on frontier science and coding benchmarks, now packaged as GGUF for local inference."
url: "https://alphasignal.ai/news/sjtu-s-bigbang-v1-beats-a-1-6t-model-using-only-machine-generated-training-data"
author: "AlphaSignal Newsroom"
author_url: "https://alphasignal.ai/editorial-team"
company: "AlphaSignal"
company_url: "https://alphasignal.ai/company/alphasignal"
published: "2026-08-07T16:29:48.000Z"
updated: "2026-09-07T09:59:33.408Z"
type: "Model"
topics:
  - "Llms"
  - "Gpus"
subtopics:
  - "Vision Language"
  - "Small Models"
  - "Quantization"
upvotes: 3500
source_url: "https://huggingface.co/bartowski/endless-frontier_BigBang-v1-GGUF"
comments: 0
truncated: true
---

# SJTU's BigBang-v1 Beats a 1.6T Model Using Only Machine-Generated Training Data

> A 35B open model post-trained on self-generated adversarial tasks matches trillion-parameter systems on frontier science and coding benchmarks, now packaged as GGUF for local inference.

BigBang-v1, a 35B model from the Endless Frontier team at Shanghai Jiao Tong University's AI Institute, landed on Hugging Face claiming scores at or above trillion-parameter frontier models on scientific research, coding, and long-horizon agent benchmarks. Bartowski followed with a full set of [GGUF quantizations](https://huggingface.co/bartowski/endless-frontier%5FBigBang-v1-GGUF) runnable on a single workstation, and the community has pulled it hundreds of thousands of times. The model's training method is what makes the benchmark numbers worth taking seriously: every training task was generated, filtered, and iterated by AI, with no human-written data in the loop.

## 35B beating 1.6T on frontier tasks

BigBang-v1 is a 35.1 billion parameter mixture-of-experts model, post-trained from Qwen3.6-35B-A3B. According to the team's [GitHub README](https://github.com/endless-frontier/BigBang-v1), it achieves the highest reported score among 35B-scale models across eight benchmarks and exceeds DeepSeek V4 Pro Preview (1.6T) on FrontierScience Research, Humanity's Last Exam, PaperBench (Code-Dev), and BioMysteryBench-HD. The evaluation suite spans long-horizon search, software engineering, scientific research, and AI research tasks. Default context length is 262,144 tokens.

_Preview only. Full text requires an AlphaSignal Pro account._

## Sources

- [bartowski/endless-frontier_BigBang-v1-GGUF](https://huggingface.co/bartowski/endless-frontier_BigBang-v1-GGUF)
- [endless-frontier/BigBang-v1](https://huggingface.co/endless-frontier/BigBang-v1)
- [BigBang-v1 GitHub repository](https://github.com/endless-frontier/BigBang-v1)
- [BigBang-v1 on Featherless AI](https://featherless.ai/models/endless-frontier/BigBang-v1)
- [Shanghai Jiao Tong Team's BigBang-V1 coverage](https://picx.dev/news/I6iboP)