As Large Language Models (LLMs) approach human expert performance, their continued development is increasingly constrained by training tasks conceived within the limits of human knowledge. We argue that open-ended capability growth requires verifiable frontier tasks: problems at the boundary of current knowledge whose solutions can be objectively evaluated through formal methods, computation, simulation, or domain-specific tools. To this end, we introduce BigBang, a general-purpose LLM evolved from Qwen 3.6 35B-A3B through efficient post-training with an adversarial, self-evolving synthetic data framework. The framework contains two core components: (1) generator agents that continually propose and solve increasingly challenging scientific and technical problems, and (2) critic agents that evaluate correctness, difficulty, scalability, and diversity, while using held-out real research tasks to calibrate the evolving synthetic-data distribution. Through iterative generator–critic interaction, the framework constructs approximately 10,000 high-difficulty post-training examples across multiple domains. Despite the modest data scale, BigBang substantially outperforms its base model across scientific research, reasoning, coding, and tool-use benchmarks, achieving aggregate performance between DeepSeek V4 Flash (284B) and DeepSeek V4 Pro (1.6T). These results demonstrate that self-evolving synthesis of verifiable frontier tasks provides a promising path toward scalable and open-ended intelligence.
Main Results
BigBang-V1 on eight representative benchmarks spanning long-horizon search, software engineering, scientific research, and AI research. BigBang-V1 obtains the highest reported score among the selected 35B models on all eight benchmarks. It even exceeds DeepSeek V4 Pro Preview (1.6T) on FrontierScience Research, Humanity's Last Exam, PaperBench(Code-Dev) and BioMysteryBench-HD.
BigBang-V1 on eight representative benchmarks. Click the figure to view the PDF.
Benchmark Results
Comparison of BigBang-V1 with representative closed- and open-source frontier models, together with models at the 35B scale, across benchmarks for long-horizon search, coding, scientific research, and AI research. The "-" indicates the score is not publicly available or not tested.
Benchmark
Claude Opus 4.8
Gemini 3.1 Pro
GPT 5.5
GLM 5.2
DeepSeek V4 Flash Preview
DeepSeek V4 Pro Preview
Step-3.7 Flash
Qwen3.6 35B-A3B
Nex-N2 mini
Agents A1
Apodex 1.0-mini
BigBang V1
Long-horizon Search
BrowseComp
84.3
85.9
84.4
68.7
73.2
83.4
75.8
67.9
74.1
48.5
73.9
76.5
XBench
61.4
-
72.4
65.8
62.2
64.8
50.8
32.6
57.2
52.4
61.8
58.4
Coding Tasks
SWE-Bench Pro
69.2
54.2
58.6
62.1
52.6
55.4
56.3
43.6
50.2
42.3
38.7
54.2
SciCode-V-Sub
92.3
-
95.1
84.3
83.7
90.2
-
56.5
39.0
64.1
-
68.6
SciCode-V-Main
78.1
-
90.6
70.3
68.6
78.1
-
26.6
15.6
50.0
-
50.0
Scientific Research
FS-R
45.2
24.8
58.3
52.4
37.7
40.7
37.2
11.9
36.8
38.4
29.6
46.2
HLE
57.9
51.4
52.2
54.7
45.1
48.2
47.2
36.2
38.4
46.3
45.3
50.3
BioMystery-HS
88.5
-
76.7
75.3
68.0
64.4
57.5
44.8
42.9
48.9
50.2
57.5
BioMystery-HD
42.4
-
23.5
21.6
23.5
13.7
11.8
2.0
5.9
2.0
5.9
15.7
AI Research
MLE-Bench(Lite)
63.6
-
59.1
72.7
40.9
59.1
40.9
31.8
18.2
27.3
27.3
59.1
PaperBench(Code-Dev)
-
-
64.2
63.6
40.4
50.4
36.7
30.7
14.8
17.3
20.5
53.6
Quickstart
For streamlined integration, we recommend using BigBang-V1 via APIs. Below is a guide to use BigBang-V1 via OpenAI-compatible API.
Serving BigBang-V1
BigBang-V1 can be served via APIs with popular inference frameworks.
In the following, we show example commands to launch OpenAI-Compatible API servers for BigBang-V1 models.
[!Important]
Inference efficiency and throughput vary significantly across frameworks.
We recommend using the latest framework versions to ensure optimal performance and compatibility.
For production workloads or high-throughput scenarios, dedicated serving engines such as SGLang, KTransformers or vLLM are strongly recommended.
[!Important]
The model has a default context length of 262,144 tokens.
If you encounter out-of-memory (OOM) errors, consider reducing the context window.
However, because BigBang-V1 leverages extended context for complex tasks, we advise maintaining a context length of at least 128K tokens to preserve thinking capabilities.
SGLang
SGLang is a fast serving framework for large language models and vision language models.
sglang>=0.5.10 is recommended for BigBang-V1, which can be installed using the following command in a fresh environment:
The following will create API endpoints at http://localhost:8000/v1:
Standard Version: The following command can be used to create an API endpoint with maximum context length 262,144 tokens using tensor parallel on 8 GPUs.
vLLM is a high-throughput and memory-efficient inference and serving engine for LLMs.
vllm>=0.19.0 is recommended for BigBang-V1, which can be installed using the following command in a fresh environment:
The following will create API endpoints at http://localhost:8000/v1:
Standard Version: The following command can be used to create an API endpoint with maximum context length 262,144 tokens using tensor parallel on 8 GPUs.
KTransformers is a flexible framework for experiencing cutting-edge LLM inference optimizations with CPU-GPU heterogeneous computing.
For running BigBang-V1 with KTransformers, see the KTransformers Deployment Guide.
Hugging Face Transformers
Hugging Face Transformers contains a lightweight server which can be used for quick testing and moderate load deployment.
The latest transformers is required for BigBang-V1:
pip install "transformers[serving]"
See its documentation for more details. Please also make sure torchvision and pillow are installed.
Then, run transformers serve to launch a server with API endpoints at http://localhost:8000/v1; it will place the model on accelerators if available:
