InternScience/Agents-A1 · Hugging Face
Summary
Agents-A1 is a 35B Mixture-of-Experts agentic model from InternScience that achieves competitive performance against frontier-scale systems like GPT-5.5 and DeepSeek-V4-pro using long-horizon trajectory scaling and multi-teacher multi-domain distillation.
View Cached Full Text
Cached at: 06/30/26, 07:39 AM
InternScience/Agents-A1 · Hugging Face
Source: https://huggingface.co/InternScience/Agents-A1
https://huggingface.co/InternScience/Agents-A1#agents-a1-scaling-the-horizon-not-the-parameters-reaching-trillion-parameter-performance-with-a-35b-agentAgents-A1: Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent
This repository contains model weights and configuration files for Agents-A1 in the Hugging Face Transformers format. These artifacts are compatible with Hugging Face Transformers, vLLM, SGLang, etc.
Agents‑A1is a 35B Mixture‑of‑Experts agentic model fromInternScience, built to scale heterogeneous agentic abilities across multiple domains includingLong‑horizon Search, Engineering, Scientific Research, Instruction Following, and Tool-calling. We investigate agent-horizon scaling from two perspectives: scaling long-horizon trajectories and scaling heterogeneous agent abilities.
From the scaling of long-horizon trajectories,Agents‑A1is trained with the assistance of a domain-grounded knowledge-action infrastructure that jointly constructs actions, observations, and verifier outcomes, turning the agent’s process into a trainable target. From the scaling of heterogeneous agent abilities,Agents‑A1presents a three-stage training paradigm for building scalable general-purpose agentic model. First, we perform full-domain supervised fine-tuning to align the base model with broad agentic behaviors. Second, we train domain-level teacher models to capture specialized expertise in each domain. Third, we propose multi-teacher multi-domain on-policy distillation with heterogeneity-aware optimization to improve knowledge transfer efficiency across different domains.
https://huggingface.co/InternScience/Agents-A1#highlightsHighlights
- Agentic Reasoning: Agents-A1 excels at decomposing complex tasks into executable sub-steps, planning ahead, and adapting its strategy based on intermediate results.
- Tool Use: Natively supports function calling and tool integration, enabling seamless interaction with APIs, code interpreters, search engines, and other external tools.
- Scientific and Professional Reasoning: Handles tool-integrated scientific reasoning and professional knowledge question answering.
- Instruction Following: Precisely follows detailed, multi-constraint instructions across diverse domains.
We welcome developers and enterprises to integrate and try Agents-A1 and share their feedback.
https://huggingface.co/InternScience/Agents-A1#performancePerformance
We evaluate Agents-A1 in real-world agentic and research-oriented workflows across six directions — long-horizon search, engineering tasks, scientific research, instruction following, general agentic tasks, and scientific agentic tasks. Despite operating in the ~35B model class, Agents-A1 delivers highly competitive performance against frontier-scale systems such as GPT-5.5, DeepSeek-V4-pro, and Kimi-K2.6. It achieves overall SOTA results on several challenging benchmarks, including Seal-0 (56.4), HiPhO (46.4), FrontierScience-Olympiad (79.0), FrontierScience-Research (40.00), IFBench (80.6), and IFEval (94.8), while also ranking as the best among comparable models on a broad range of tasks such as BrowseComp (75.5), XBench-DS-2510 (86.0), GAIA (96.0), SciCode (44.3), HLE with tools (47.6), and MolBench-bind (56.8). These results show that Agents-A1 combines strong long-horizon search ability, robust scientific reasoning, and reliable instruction following, establishing it as a highly capable and efficient agentic model that narrows the gap with much larger frontier models.
🥇 Overall SOTA 🟢 Best Among Comparable Models (~35B)
Benchmark📏 Comparable Models (~35B)🚀 Larger-scale Models⭐ OursQwen3.5-35B-A3BQwen3.6-35B-A3BNex-N2-miniStep-3.5-FlashKimi-K2.6DeepSeek-V4-pro(Max)GPT-5.5(xhigh)Agents-A1**🔍 Long-horizon SearchBrowseComp61.067.9374.169.083.283.4🥇 84.4🟢 75.51XBench-DS-251077.071.082.056.3🥇 90.0🥇 90.084.0🟢 86.0Seal041.438.7449.5536.9450.4554.9542.34🥇 56.36GAIA59.878.6482.5284.580.58🥇 98.0687.38🟢 96.04⚙️ Engineering TasksSciCode37.735.829.940.453.550.0🥇 56.1🟢 44.33MLE-Lite24.2434.8534.8554.5562.1263.64🥇 72.73🟢 43.94🧪 Scientific ResearchHLE w/ tools47.436.232.023.1🥇 54.048.252.2🟢 47.6HiPhO37.037.738.538.341.138.743.3🥇 46.4FrontierScience-Olympiad64.560.352.061.073.076.078.0🥇 79.0FrontierScience-Research2.52.95.06.717.913.326.7🥇 40.0📋 Instruction FollowingIFBench70.264.454.0864.671.7773.4775.9🥇 80.61LongBench-v259.057.759.657.562.0🥇 64.3-🟢 60.2IFEval91.991.388.493.5394.4593.3593.35🥇 94.82🤖 General Agentic Tasksτ2-Bench🟢 81.279.074.5375.7781.93🥇 82.281.6379.81VitaBench31.935.623.030.035.63🥇 49.0445.0🟢 38.75🔬 Scientific Agentic Tasks**MatTools21.015.934.144.9363.847.1🥇 68.8🟢 47.1MolBench-bind46.048.751.445.9521.637.8🥇 62.2🟢 56.8## https://huggingface.co/InternScience/Agents-A1#usageUsage
https://huggingface.co/InternScience/Agents-A1#sglangSGLang
SGLangis a fast serving framework for large language models and vision language models.
Install SGLang with uv:
uv venv --python 3.12 --seed --managed-python
source .venv/bin/activate
uv pip install sglang
Seeits documentationfor more details.
The following commands create API endpoints athttp://localhost:8000/v1:
- Standard Version(1 GPUs, 262K context):
python -m sglang.launch_server \ --model-path InternScience/Agents-A1 \ --port 8000 \ --tp-size 1 \ --mem-fraction-static 0.8 \ --context-length 262144 \ --reasoning-parser qwen3 - Tool Use:
python -m sglang.launch_server \ --model-path InternScience/Agents-A1 \ --port 8000 \ --tp-size 1 \ --mem-fraction-static 0.8 \ --context-length 262144 \ --reasoning-parser qwen3 \ --tool-call-parser qwen3_coder
https://huggingface.co/InternScience/Agents-A1#vllmvLLM
vLLMis a high-throughput and memory-efficient inference and serving engine for LLMs.
Install vLLM from the main branch via uv:
uv venv --python 3.12 --seed --managed-python
source .venv/bin/activate
uv pip install vllm --torch-backend=auto
Seeits documentationfor more details.
The following commands create API endpoints athttp://localhost:8000/v1:
- Standard Version(1 GPUs, 262K context):
vllm serve InternScience/Agents-A1 \ --port 8000 \ --tensor-parallel-size 1 \ --max-model-len 262144 \ --reasoning-parser qwen3 - Tool Call:
vllm serve InternScience/Agents-A1 \ --port 8000 \ --tensor-parallel-size 1 \ --max-model-len 262144 \ --reasoning-parser qwen3 \ --enable-auto-tool-choice \ --tool-call-parser qwen3_coder - Text-Only(skips vision encoder to free KV cache memory):
vllm serve InternScience/Agents-A1 \ --port 8000 \ --tensor-parallel-size 1 \ --max-model-len 262144 \ --reasoning-parser qwen3 \ --language-model-only
https://huggingface.co/InternScience/Agents-A1#recommended-sampling-parametersRecommended Sampling Parameters
For the best generation quality, we recommend the following sampling parameters:
temperature: 0.85top\_p: 0.95top\_k: 20min\_p: 0.0presence\_penalty: 1.1repetition\_penalty: 1.0
https://huggingface.co/InternScience/Agents-A1#agent-capability-evaluationAgent Capability Evaluation
To provide the community with a unified agent evaluation codebase for fair comparison, we have also open-sourced an evaluation framework for assessing agentic models across core capabilities, including tool use and multi-step reasoning. The evaluation code is included in theAgents-A1/evaluationof this repository.
We use this framework to evaluate the released model under a standardized and reproducible setting. Specifically, the model is tested on a set of agent-oriented tasks that require it to understand user goals, decompose complex instructions, interact with tools or environments when necessary, and produce final results. The evaluation results reported inModel Cardare generated using the open-source framework above, so that users can reproduce the experiments, compare other models under the same protocol, and further extend the benchmark for new agent scenarios. (**Note that:**To ensure a fair comparison, we report the benchmark results from their original technical reports. If a model does not report the corresponding benchmark results, we evaluate it using the same evaluation protocol as our model.)
For detailed evaluation scripts, task definitions, metrics, and reproduction instructions, please refer to the evaluation codebase.
https://huggingface.co/InternScience/Agents-A1#citationCitation
If you find our work helpful, feel free to give us a cite.
Similar Articles
Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent
Introduces Agents-A1, a 35B Mixture-of-Experts agentic model that achieves trillion-parameter-level performance through long-horizon trajectory scaling and a three-stage training approach including SFT, domain-level teachers, and multi-teacher distillation. The model outperforms or matches much larger models on long-horizon agent benchmarks.
@ModelScope2022: Introducing Agents-A1, A 35B MoE agentic model built for long-horizon tasks across search, engineering, scientific rese…
ModelScope introduces Agents-A1, a 35B MoE agentic model with 256K context and function calling, achieving SOTA on long-horizon tasks and instruction following.
Agent S2: A Compositional Generalist-Specialist Framework for Computer Use Agents
Agent S2 is a new compositional framework for computer use agents that achieves state-of-the-art performance on multiple benchmarks by utilizing Mixture-of-Grounding and Proactive Hierarchical Planning.
Agent-World: Scaling Real-World Environment Synthesis for Evolving General Agent Intelligence
Agent-World introduces a self-evolving training framework for general agent intelligence that autonomously discovers real-world environments and tasks via the Model Context Protocol, enabling continuous learning. Agent-World-8B and 14B models outperform strong proprietary models across 23 challenging agent benchmarks.
@AnandButani: ml-intern by @huggingface is wild You drop a high-level prompt (“build the best scientific reasoning model” or “crush h…
Hugging Face’s open-source "ml-intern" agent automates the full post-training pipeline—from literature review and data cleaning to model tuning—given only a high-level prompt.