LLM Reference

Agents-A1

Released
2026-06-26
Last refreshed
2026-07-07
Status
Researched today
Open sourceCommercial use: permittedMultimodalRAGAgentsLong contextVisionJSON / Tool useAgentsVision

Agents-A1 is a released rag, agents, and long context model with open-source and 262k context; evaluate it while provider pricing coverage matures.

Use it for

  • Teams evaluating rag, agents, and long context
  • Workloads that can use a 262k context window

Do not use it for

  • Cost-sensitive launches that need sourced token pricing
  • Teams that need a tracked hosted API route today
Specifications
Family
Agents-A1
Released
2026-06-26
Context
262k
Parameters
35B total, 3B active
Architecture
Mixture of Experts
Specialization
general
Openness
Open source
License
Apache 2.0OSI-approvedCommercial use: permitted
Weights
Available
Code
Unknown
Training
Multi-stage
Created by

AI for Science Center at Shanghai AI Laboratory.

Shanghai, China
Website
Pricing

No tracked provider token pricing is available yet.

About

Agents-A1 is Intern Science's 35B-A3B open-source Mixture-of-Experts agentic model for long-horizon search, engineering, scientific research, instruction following, and tool calling. It is trained with a three-stage recipe that combines full-domain supervised fine-tuning, domain-level teacher models, and multi-teacher on-policy distillation over knowledge-action trajectories. The model supports vision-language inputs, reasoning traces, and tool/function calling through standard LLM runtimes such as SGLang and vLLM, with a recommended served context length of 262,144 tokens. Intern Science reports strong vendor scores on BrowseComp, HLE with tools, IFEval, and scientific/engineering agent benchmarks; teams should treat these as model-card scores until independently reproduced.

Agents-A1 is Intern Science's 35B-A3B open-source Mixture-of-Experts agentic model for long-horizon search, engineering, scientific research, instruction following, and tool calling. It is trained with a three-stage recipe that combines full-domain supervised fine-tuning, domain-level teacher models, and multi-teacher on-policy distillation over knowledge-action trajectories. The model supports vision-language inputs, reasoning traces, and tool/function calling through standard LLM runtimes such as SGLang and vLLM, with a recommended served context length of 262,144 tokens. Intern Science reports strong vendor scores on BrowseComp, HLE with tools, IFEval, and scientific/engineering agent benchmarks; teams should treat these as model-card scores until independently reproduced.

Agents-A1 has a 262k-token context window.

Top use-case fit: coding, agents, and build tasks

RAG

Included by capability and metadata signals in the decision map.

Agents

Included by capability and metadata signals in the decision map.

Long context

Included by capability and metadata signals in the decision map.

Provider price ladder

No tracked provider token pricing is available for this model yet.

Capabilities

VisionMultimodalReasoningFunction CallingTool Use

Benchmark peer barsfor RAG

No task-mapped benchmark peers are available for this model yet.

Benchmark scores(3)

Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.
BenchmarkScoreVersionSource
BrowseComp75.5BrowseComp accuracy, vendor-reported on Agents-A1 model cardhttps://huggingface.co/InternScience/Agents-A1
Humanity's Last Exam — With Tools47.6Humanity's Last Exam with tools accuracy, vendor-reported on Agents-A1 model cardhttps://huggingface.co/InternScience/Agents-A1
Instruction-Following Evaluation94.8IFEval accuracy, vendor-reported on Agents-A1 model cardhttps://huggingface.co/InternScience/Agents-A1

Migration checks

No linked migration route is available for this model yet.

API versions

InternScience/Agents-A1InternScience/Agents-A1-FP8InternScience/Agents-A1-Q4_K_M-GGUFInternScience/Agents-A1-Q8_0-GGUFInternScience/Agents-A1-F16-GGUF

Frequently asked questions

What is the context window of Agents-A1?

Agents-A1 has a context window of 262k tokens.

When was Agents-A1 released?

Agents-A1 was released on 2026-06-26.

What benchmarks has Agents-A1 been tested on?

Agents-A1 has been evaluated on 3 benchmarks, including BrowseComp, Humanity's Last Exam — With Tools, Instruction-Following Evaluation.