Apodex launches 1.1, open-sources FrontierAgent framework and 35B mini model

Apodex released version 1.1 alongside the open-source agent framework FrontierAgent and the 35B-parameter 1.1 mini model. The release rests on one thesis: the meaningful unit of AI capability is end-to-end completion of real work with verifiable deliverables, not merely answering questions — "a correct answer can still be a failed task." With its generation/verification split and low-barrier open-source combination, it is worth watching for anyone concerned with trustworthy agent output.

Confirmed

• FrontierAgent is open-source on GitHub (2.4k stars); one command starts it on macOS/Linux with no prerequisites and no hard Docker dependency; native CLI TUI with single-agent and Agent Team modes
• Apodex 1.1 mini is a 35B open-weight model on Hugging Face with FP8, NVFP4, GPTQ-Int4 quantizations, usable with FrontierAgent for local deployment
• The full model powers the online web workspace, with API free for two weeks for new users
• Official numbers place 1.1 with Agent Team in the current frontier tier on APEX-Agents, GDPval, FrontierScience and Humanity's Last Exam; per a repost, the Apodex 1.1 paper (arXiv:2608.23283) and Agent Team results are indexed on Papers with Code, with a 63.3% FrontierScience pass rate
• Statement Review separates generation from checking: key conclusions are independently reviewed before delivery, flagging weak evidence, mismatched citations and calculation conflicts while retaining an audit trail
• Deep Discover mode runs an async Agent Team: the model decides task splitting, sub-agent count and when to consolidate; new files or changed requirements can be injected mid-run, preserving valid work and re-planning only affected parts — official demo is a global EV battery supplier risk assessment
• Official demos handle real professional files, e.g. cleaning raw clinical data and producing Kaplan-Meier survival curves, plus powder XRD indexing and batch protein docking

Why it matters

• The release shifts from "telling you how" to "working alongside you," making verifiable delivery rather than Q&A accuracy the capability standard — directly targeting trustworthiness of agent output in real professional settings
• Reviewer aakashgupta considers the generation/checking split of Statement Review the most underrated capability and a key safeguard for using agent output in real work
• The open-source framework plus open 35B weights lower the barrier to local reproduction, with mini running full workflows locally alongside FrontierAgent

2026-09-08 ~ 2026-09-09 · 14 related posts

• Apodex 1.1 Agent Team hits 63.3% pass rate on FrontierScience-Research, paper lands on Papers with Code — NielsRogge · 2026-09-08
• [source] Apodex 1.1 Argues a Right Answer Can Still Be a Failed Task — aakashgupta · 2026-09-09
• Apodex 1.1 argues a right answer can still be a failed task — aakashgupta · 2026-09-09
• Inside Apodex: an asynchronous Agent Team running on a live task board — aakashgupta · 2026-09-09
• Apodex agents absorb mid-task requirement changes with a partial replan — aakashgupta · 2026-09-09
• Apodex 1.1 review: Statement Review independently verifies agent conclusions — aakashgupta · 2026-09-09
• Apodex 1.1 Demos Hit Real Clinical Data, and Statement Review May Be the Sleeper Feature — aakashgupta · 2026-09-09
• Apodex Claims 1.1 With Agent Team Sits in the Frontier Tier; 35B Mini Runs Locally — aakashgupta · 2026-09-09
• [source] Apodex 1.1 Mini Weights Hit Hugging Face; Full Model API Free for Two Weeks — aakashgupta · 2026-09-09
• Apodex Open-Sources FrontierAgent: One-Command Local Agent Framework With Agent Team Mode — aakashgupta · 2026-09-09
• FrontierAgent splits generation from verification, flagging weak evidence before delivery — mhdfaran · 2026-09-09
• Apodex open-sources FrontierAgent: local multi-agent TUI framework with ReAct and Agent Team modes — mhdfaran · 2026-09-09
• [source] Apodex open-sources FrontierAgent agent framework and ships 35B open-weight Apodex 1.1 mini — mhdfaran · 2026-09-09
• Apodex 1.1 launches with open-weight 35B mini model for local deployment — mhdfaran · 2026-09-09