Nanjing University

Headlines · 3

1. 6d
   OpenMobile-2 releases open smartphone-agent models and training environment for more than 110 apps

   Researchers from Nanjing University, SenseTime, and other universities released OpenMobile-2, including 9B and 27B models built on Qwen3.5 and Qwen3.6, together with the training environment, data, and reproduction instructions. The environment covers more than 110 simulated applications. The researchers reported that, with the same amount of training data, increasing app diversity raised success on unseen applications from 60.7% to 67.5%. They also reported that the 27B model scored 59.6 on SPA-Bench, the highest result among open models. These findings come from one secondary report and remain researcher-reported results.

   The Decoder · 🔥 0
2. Oct 19
   Study finds dynamic concurrency in coding agents helps long tasks but hurts short ones

   Researchers from Nanjing University and UIUC ran paired experiments on the dynamic concurrency mechanisms of three coding agents — Codex, Claude Code and Kimi Code — across 354 software engineering tasks and 2,124 executions, analysing full main-agent and sub-agent traces. Results varied by task type and system: on LoopsBench, Claude Code's completion rate rose from 7.1% sequentially to 21.4% with concurrency, and Codex rose from 34.6% to 57.7% on NL2Repo; on SWE-bench Verified, Claude Code fell from 83.0% to 59.0% and Kimi Code from 77.0% to 59.0%, while Codex stayed at 71.0% in both modes. Trace analysis produced a taxonomy of 4 categories, 13 subcategories and 28 concurrency failure modes (such as missing owners, pseudo-concurrency, load imbalance, mistimed result merging and shared-environment conflicts), plus three mechanisms where concurrency paid off: complementary work, parallel exploration and independent verification.

   新智元 · 🔥 0
3. Oct 16
   SinapisAI's Mio 1.0 Lite tops nine benchmarks at lowest cost, vendor report says

   Machine Heart reports that SinapisAI, a startup incubated at Nanjing University, published a technical report titled "Toward Collective Model Intelligence," disclosing for the first time its framework for coordinating large numbers of models: quantified capability profiles for each model, a Mio router that picks and combines models based on task requirements and budget, and an underlying Model Space that manages versions and onboarding for 484 models. The report says Mio 1.0 Lite ranked first with a 77.15/100 average across nine public benchmarks at the lowest average cost of 0.06 yuan per request, and took top scores on five of the datasets; the figures are self-published and the account shows no independent reproduction. The report also acknowledges that cost-no-object flagship models retain a quality edge on hard tasks such as HLE, and that the overall benefit of integrating the system, along with the vision of collective intelligence improving with experience, still needs verification.

   机器之心 · 🔥 0

Experience and discussion from the community

Share my Nanjing University experienceAsk about Nanjing University