ai-trend-notifier
← wiki

$ cat wiki/papers/2026/2606.30616-agents-a1.md

Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent

paperupdated 2026-07-01created 2026-07-01

TL;DR

A 35B MoE model (Agents-A1) matches 1-trillion-parameter models on agentic benchmarks by scaling the agent horizon (trajectory length + ability breadth) rather than raw parameter count. Three-stage training: full-domain SFT → domain-level teacher models → multi-teacher routed on-policy distillation.

Authors & Org

InternScience team, Shanghai Jiao Tong University (SJTU). Weights released publicly on HuggingFace Hub: InternScience/Agents-A1. arXiv: 2606.30616 GitHub: https://github.com/InternScience/Agents-A1

Method

Insight: agentic performance bottleneck is the agent horizon (how long and how broadly an agent can act), not model size.

Infrastructure: a long-horizon knowledge-action chain that interleaves external knowledge retrieval, tool actions, observations, and verifier feedback → average trajectory length 45K tokens (vs. typical <8K for non-agentic tasks).

Training recipe:

  1. Full-domain SFT — broad agentic behavioral alignment on the base MoE
  2. Domain-level teacher model training — specialized teacher per domain (genomics, math, coding, chemistry, etc.) capturing domain expertise
  3. Multi-teacher domain-routed on-policy distillation with salient vocabulary alignment — transfers domain-specific knowledge efficiently; routing ensures each input is distilled by the most relevant teacher

Results

BenchmarkAgents-A1 (35B MoE)Note
SEAL-056.4Leading vs. 1T models
IFBench80.6Leading
HiPhO46.4Leading
FrontierScience-Olympiad79.0Leading
MolBench-Bind56.8Leading
SciCode44.3Highly competitive
HLE47.6Highly competitive
BrowseComp75.5Highly competitive
Compared baselines: Kimi-K2.6, DeepSeek-V4-pro (both ~1T parameters).

Significance

  1. Efficiency argument: a 35B MoE (deployable on a high-end workstation) matches 1T-parameter performance — a ~30× efficiency gain on agentic tasks specifically.
  2. Scaling law for agents: the primary scaling axis in agents may be trajectory length and ability breadth, not raw parameters. Complements the "test-time compute" narrative but applied to agent infrastructure, not just reasoning depth.
  3. Open weights: availability on HuggingFace makes this directly usable for the open-source agent ecosystem.
  4. Multi-teacher distillation: may generalize as a training recipe for efficient specialization across domains.

Open Questions

  • Does the result hold at longer horizons (>100K tokens)?
  • Is horizon-scaling a unique advantage of MoE (routing per domain) or does it transfer to dense models?
  • How does Agents-A1 perform on real-world deployment tasks vs. benchmarks?

Cite

@article{agents-a1-2026,
  title={Scaling the Horizon, Not the Parameters: Reaching Trillion-Parameter Performance with a 35B Agent},
  author={InternScience},
  journal={arXiv preprint arXiv:2606.30616},
  year={2026}
}

Related

Referenced by

Sources