Apodex 1.1 trains AI agents to deliver verifiable completed work, not polished answers
rohanpaul_ai · x · 2026-09-04
Rohan Paul breaks down Apodex 1.1 from Apodex AI: instead of judging a single polished answer, it trains the model to take over an entire job — keeping a real workspace alive across hundreds of steps, recovering from failures, and producing verifiable artifacts.
- Tasks can start from raw files, datasets, spreadsheets, papers, images, or code; the system inspects them, picks a method, runs code, maintains task state, and recovers from failures.
- Two scaling paths: Environment Scaling expands the executable file/search/code environments the model learns from; Agentic Coordination Scaling trains it to split long tasks across agents, merge partial results into shared state, and revise plans as evidence changes.
- AgentOS keeps files, tool state, artifacts, dependencies, and coordination state persistent; subagents can return useful results before every branch finishes, letting the main agent redirect unfinished work without discarding valid progress.
- Officially published results place Apodex 1.1 with Agent Team in the leading band across professional work, finance, science, reasoning, and search.
Related event: Apodex 1.1 trains agents on verifiable units of complete work(2 posts)→
More from coding & agent
- Lab lessons from Anthropic MHS: keep fast control out of the model — Empty-Abalone-2952 · 2026-09-04
- MCP veteran launches TDQS, an open spec scoring 15,000+ tool definitions — punkpeye · 2026-09-04
- WebMCP Computer: one URL gives any coding agent a disposable OS in the browser — prd_008 · 2026-09-04
- Grok Bot Hands Out 50 x $200 Codes as User Shares Orchestrator-Bot Workflow — omarsar0 · 2026-09-04
- Dev Spent 5 Months of Claude Max Improving His Open-Source App Store Connect CLI — rudrank · 2026-09-04
- WHALE: a simple recipe to jointly optimize an LLM's weights and harness — lateinteraction · 2026-09-04