SA
Shanghai AI Lab
Language model · China ·Jul 2026

Agents-A1

Language Open Source
35B
Parameters
256K
Context

About Agents-A1

Agents-A1 is an AI model developed by Shanghai AI Lab, in the language category, released in 2026, made available as an open-source model with 35B params. Shanghai AI Lab's Agents-A1 is a 35B MoE built on Qwen3.5-35B-A3B by the InternScience team, trained specifically for long-horizon agent work with a 256K context window. Shipping with quantized variants under Apache 2.0.

On this page you'll find Agents-A1's full specifications, including a context window of 256K tokens. Review provider pricing and benchmark scores below, or compare Agents-A1 head-to-head with other language models.

→ More models by Shanghai AI Lab → All language models → AI benchmarks leaderboard

Run Agents-A1 locally — what to buy

Share this build ↗

Sized to what Agents-A1 actually needs (~256k context at Q4), not to the biggest GPU: Good = cheapest that runs it, Better = best value, Best = most headroom. Speed is a hardware estimate; anything past ~120 tok/s is shown as instant.

Good
AMD Radeon RX 7900 XTX
Consumer GPU · ~$999
~37.7 tok/s
Q4 · tight on context · up to 4k ctx
Runs with: ExLlamaV2, KoboldCpp, LM Studio
See full estimate →
Better
NVIDIA GeForce RTX 4090
Consumer GPU · ~$1,599
~44.7 tok/s
Q4 · tight on context · up to 4k ctx
Runs with: ExLlamaV2, KoboldCpp, LM Studio
See full estimate →
Best
NVIDIA GeForce RTX 5090
Consumer GPU · ~$1,999
~79.5 tok/s
Q4 · fits full context · up to 32k ctx
Runs with: ExLlamaV2, KoboldCpp, LM Studio
See full estimate →

Software support

·ExLlamaV2 ·KoboldCpp ·LM Studio ·MLX ·Ollama ·SGLang ·TensorRT-LLM ·llama.cpp ·vLLM ?KTransformers ?text-generation-webui

✓ measured · · compatible · — not supported. Informational only — speed is hardware-based.

Consumer GPU
10.1–120+ tok/s
15 options →
Apple
10.3–33.5 tok/s
20 options →
Datacenter GPU
33.9–120+ tok/s
13 options →

Community

Would you run Agents-A1 again?
No votes yet — be the first.
Did it run? — community reports by hardware

No reports yet. Own this model on some hardware? Be the first to confirm it runs.

Discussion

No comments yet. Start the discussion.

More models like Agents-A1

AI
DMind-1
 AI
COLLECTIVE-1
 AI
Cogito 1 series
 AI
Euler (Euler大模型)
· 2.5B
 AI
SmolLM-1.7B
· 1.7B
 AI
Llama-3-70B-Synthia-v3.5
· 70B

Frequently asked questions

What is Agents-A1? ▶

Shanghai AI Lab's Agents-A1 is a 35B MoE built on Qwen3.5-35B-A3B by the InternScience team, trained specifically for long-horizon agent work with a 256K context window. Shipping with quantized variants under Apache 2.0. It is tracked on GenAIList with its specifications, benchmark scores and provider pricing.

Who created Agents-A1? ▶

Agents-A1 was developed by Shanghai AI Lab and released in 2026.

Is Agents-A1 open source or proprietary? ▶

Agents-A1 is released as open source — its weights are publicly available to download, run locally and fine-tune.

What is Agents-A1's context window? ▶

Agents-A1 supports a context window of 256K tokens, and has 35B params.

How much does Agents-A1 cost? ▶

Pricing for Agents-A1 depends on the provider. See the providers table on this page for the latest API rates.

How does Agents-A1 perform on benchmarks? ▶

Benchmark scores for Agents-A1 are listed on this page as they are published. You can compare it head-to-head with other models on the GenAIList compare tool.

Agents-A1: Scaling the Horizon, Not the Parameters - Test Locally
Agents-A1: a 35B agent rivals trillion-param models by scaling the horizon, not size #Shorts
Agents-A1 Reaches Trillion-Parameter Performance With 35B Parameters #Shorts

Know a great video about Agents-A1?