InternScience

Agents-A1 35B A3B

Name: Agents-A1 35B A3B
Rating: 85 (87 reviews)
Author: InternScience

最先端

HuggingFace

43.8Kダウンロード605いいねJun 2026公開日262K トークンコンテキストApache 2.0ライセンス81 優秀品質

Agents-A1 35B A3B (35.099998474121094B parameters) requires approximately 23.2 GB of VRAM with Q4_K_M quantization. As a Mixture of Experts model with 3B active parameters, it uses less memory than its total parameter count suggests. For the best balance of quality and speed, we recommend hardware with at least 27 GB of VRAM.

はじめに

— コピー&ペーストでローカル実行

Copy-paste commands to run Agents-A1 35B A3B on your machine.

Run

docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
  --hf-repo "InternScience/Agents-A1" \
  --hf-file "Agents-A1-Q4_K_M.gguf" \
  -c 4096 -ngl 99

Quick specs

Parameters35.1B (3B active)

Architecturemoe (MoE)

Context262K tokens

Modalitytext+vision

Min RAM13.7 GB

Rec. RAM21.4 GB (Q4_K_M)

LicenseApache 2.0

FamilyAgents-A1

✓ Code✓ Reasoning

About this model

Agents-A1 is InternScience's 35B Mixture-of-Experts agentic model (~3B active per token), built to scale heterogeneous agent abilities across long-horizon search, engineering, scientific research, instruction following and tool calling. Despite the ~35B class, it posts SOTA results on Seal-0, HiPhO, FrontierScience and IFBench, competing with frontier systems many times its size.

•35B total / ~3B active — runs single-GPU at 262K context.
•SOTA on Seal-0 (56.4), HiPhO (46.4), FrontierScience-Olympiad (79.0), IFBench (80.6), IFEval (94.8).
•Competitive with GPT-5.5, DeepSeek-V4-Pro and Kimi-K2.6 on agentic benchmarks.
•Vision-capable (VLM) for GUI and document agent workflows.
•Agent-horizon scaling: long-horizon trajectories plus heterogeneous agent abilities.

Agents-A1 35B A3Bのおすすめ

RTX 5090 32GBS

32 GB

RTX PRO 4500 Blackwell 32GBS

32 GB

AMD Instinct MI100 32GBS

このモデルを実行

Agents-A1 35B A3B on RTX 5090 32GB Agents-A1 35B A3B on RTX PRO 4500 Blackwell 32GB Agents-A1 35B A3B on AMD Instinct MI100 32GB

量子化オプション

量子化レベル別VRAM推定値

No hardware detected — fit column shows raw VRAM estimates

Quant	Bits	VRAM	Quality	Fit
Q1_0_G128	1.125	5.1 GB	Very Low	—
Q2_0_G128	1.71	9.4 GB	Low	—
Q2_K	2	13.7 GB	Low	—
Q3_K_S	3	17.2 GB	Low	—
NVFP4	4	19.7 GB	Medium	—
Q4_K_M	4	21.4 GB	Medium	—
Q5_K_M	5	25.3 GB	High	—
Q6_K	6	28.8 GB	High	—
Q8_0	8	37.6 GB	Very High	—
F16	16	72.0 GB	Maximum	—

Quality benchmarks

Agents-A1 35B A3B benchmark scores

Benchmark verified

General

Chatbot Arena—

IFEval94.8%

Source: official · 2026-06-22

ハードウェア互換性

全ハードウェアの適合度推定

カリキュレーターを開く

Computing compatibility...

メモリ内訳

Reference: RTX 2060 6GB

Weights21.4 GB

KV Cache0.3 GB

Runtime0.9 GB

Headroom0.6 GB

よくある質問

Agents-A1 35B A3B

Agents-A1 35B A3Bのおすすめ

量子化レベル別VRAM推定値

Agents-A1 35B A3B benchmark scores

全ハードウェアの適合度推定

Reference: RTX 2060 6GB

FAQ — Agents-A1 35B A3B