Apodex 1.1 shifts from research Q&A to end-to-end task execution

Apodex released its 1.1 model family on Sept 16 (live since Aug 24 per TestingCatalog), pivoting from deep-research Q&A toward end-to-end execution of long-horizon real tasks, prompting multiple authors to publish teardown analyses.

Confirmed

• Per @testingcatalog, Apodex 1.1 no longer just reads webpages and returns text: it opens uploaded files (spreadsheets, PDFs, datasets), chooses analysis methods, writes and runs analysis code, and returns traceable tables, charts and scripts where every conclusion traces back to its source.
• @aakashgupta relays the core claim that a correct answer can still be a failed task: the meaningful unit of AI capability is end-to-end completion of real work — six things including understanding goals, handling files, writing code, and producing verifiable results — not just a plausible report.
• @aakashgupta tested mid-task intervention: adding new files or changing requirements mid-run preserves valid work and only replans affected parts; the official demo is a global analysis task.
• On the engine side, both @aakashgupta and @testingcatalog describe an asynchronous Agent Team running in Deep Discover mode: the model decides whether to split tasks, how many sub-agents to launch, and when to aggregate; sub-agents run in parallel, stream results into shared task state, and feature adaptive recovery.
• @testingcatalog notes Apodex 1.1 shows results on APEX-A under the Agent Team configuration, without specific numbers.

Why it matters

• The release marks a clear positioning shift from answering questions to delivering verifiable work products, aligned with industry focus on long-horizon agentic execution.
• Mid-task intervention and incremental replanning address the real pain of frequently changing requirements, a key engineering trait for practical agents.

2026-09-16 ~ 2026-09-16 · 5 related posts

• [source] Apodex 1.1 launches: moves from deep research to real task execution — testingcatalog · 2026-09-16
• [source] Inside Apodex 1.1: Adaptive Recovery, AgentOS, and self-organizing async agent teams — testingcatalog · 2026-09-16
• [source] Apodex 1.1 argues a right answer can still be a failed task—execution is the unit of AI capability — aakashgupta · 2026-09-16
• Inside Apodex 1.1: an async Agent Team in Deep Discover mode on a live task board — aakashgupta · 2026-09-16
• Apodex 1.1 lets you drop new files mid-task, replanning only the affected parts — aakashgupta · 2026-09-16