AI Coding Live News

Mathematicians working with Shanghai AI Lab's Intern-S2 resolve Ghys's conjecture in complex dimension three; four main theorems checked in Lean by up to 100 agents in ~5 days

Shanghai AI Lab announced on 2026-10-15 that Disheng Xu (Great Bay University) and Jiesong Zhang (KTH), aided by an internal version of Intern-S2, proved the codimension-one case of Ghys's 1995 holomorphic Anosov conjecture, including complex dimension three (arXiv:2609.37695). Peking University's team formalised all four main theorems in Lean with the SHEAF framework in about five days with up to 100 concurrent agents and ~550k lines, sorry-free and accepted by Lean comparator, with Hsiang–Wall (1969) as an explicit hypothesis.

Mathematicians working with Shanghai AI Lab's Intern-S2 resolve Ghys's conjecture in complex dimension three; four main theorems checked in Lean by up to 100 agents in ~5 days

Key Takeaways

  • Result: codimension-one Ghys conjecture proved — the manifold is a complex torus and the map is holomorphically conjugate to a hyperbolic torus automorphism; higher dimensions remain open (arXiv)
  • Human–AI loop: problem graph grew 15 → 389 nodes over seven rounds; ~400-page first report, ~1,000 pages total (announcement)
  • Formalisation: ~5 days, up to 100 agents, ~550k lines of Lean (team-reported); sorry-free, three standard axioms (repo)
  • Tooling: SHEAF proves top-down over a DAG; in a prior run 62.5% (1,067/1,706) of non-Mathlib nodes were pruned as unnecessary
  • Context: Intern-S1 scored 102 in the official AI evaluation at the 2025 Chinese Mathematical Olympiad finals vs a 78-point human gold threshold
Read full story

vLLM day-0 support for Intern-S2-397B: multimodal, reasoning, coding, scientific agents

vLLM announced day-0 serving support for Shanghai AI Lab Intern-S2-397B. The 397B multimodal foundation model targets long-horizon scientific research and scientific agents, with strong multimodal, reasoning, coding, and agent capabilities. Developers can run it on vLLM today via the official recipe.

Key Takeaways

  • vLLM shipped day-0 serving for Intern-S2-397B across multimodal, reasoning, coding, and scientific-agent workloads.
  • Official recipe is at recipes.vllm.ai/internlm/Intern-S2-397B for immediate deployment.
  • Pairs with the same-day Intern-S2-397B weight/eval release already in the feed: open weights plus open serving stack.
Read full story
Linked Models & Agents:
⚡ Intern S2 397b🌐 shanghai-ai-lab

Shanghai AI Lab Intern-S2-397B ships BF16/FP8: 84.0 SWE-bench Multilingual, Apache-2.0

ModelScope announced Shanghai AI Laboratory Intern-S2-397B weights in BF16 and FP8 under Apache-2.0. Reported scores include 87.0 on FrontierScience-Olympiad and 84.0 on SWE-bench Multilingual, with long-horizon agent RL, tool use, and thinking/non-thinking modes. Training ingests raw scientific pages (text, visuals, symbols) across 20+ domains. Collection on ModelScope.

Key Takeaways

  • BF16 and FP8 weights are public under Apache-2.0 on ModelScope.
  • Reported 84.0 SWE-bench Multilingual and 87.0 FrontierScience-Olympiad as comparison leaders.
  • Long-horizon agent RL targets tool use and sustained execution, with thinking and non-thinking modes.
Read full story
Linked Models & Agents: