PRODUCT SUITE
The highest-rigor heavy-duty solver for the hardest questions at the frontier of human knowledge. It explores multiple paths, generates and tests hypotheses, verifies critical claims, and converges on the strongest answer.
Heavy-Duty Multi-Agent Architecture
Advanced reasoning model with full agent-team orchestration, parallel exploration, verification, and final synthesis. Deep Discover explores multiple paths, generates and tests hypotheses, verifies critical claims, and converges on the strongest answer.
Orchestrator
Decomposes the task and spawns up to 150 sub-agents asynchronously across retrieval, synthesis, and verification roles.
Shared Evidence Pool
An append-only structured store where sub-agents deposit findings. The orchestrator queries it asynchronously, enabling 15,000+ step workflows.
Verification Team
Conflict Reviewer, Fact Checker, and Global Verifier operate independently — never sharing a context window with reasoning agents — to prevent confirmation bias.
INPUT
User Query / Research Objective
ORCHESTRATOR
Decomposes task · Spawns sub-agents · Monitor Progress · Detect Conflict · Trigger Verification
SUB-AGENT POOL - UP TO 150 ACTIVE AGENTS
Expert Agent team
Dynamically assembled to solve each problem
Retrieval
Browse, query, extract raw data from sparse sources
Synthesis
Compact retrieved info into structured evidence
Reasoning
Formulate hypotheses, evaluate alternative paths
Verification Team
Audit claims against source documents in real time
Conflict Reviewer
Resolves contradictions between sub-agents
Fact Checker
Grounds claims against raw source evidence
Draft-report Reviewer
Reasons over the full evidence graph
· Sub-agents deposit findings as they complete
· Verification agents draw from the same pool in parallel
SHARED EVIDENCE POOL - ASYNC STATUS TABLE
Evidence ID
Source
Confidence
EV-091
Academic Paper (PDF)
EV-092
Clinical Trial Registry
EV-093
Patent Database
ORCHESTRATOR
Synthesizes Final Report
OUTPUT
Verified Final Answer & Evidence Graph
How Discover Redefines the Landscape
Other Deep Research Models
Apodex Deep Research
Apodex Deep Solve
Apodex Deep Discover
Approach
Search → Summarize
Search → Synthesize
Search → Reason
Reason → Verify → Discover
Verification
Self-reflection only
Self-reflection only
Self-reflection only
In-flight team + Local & Global Verifier
Scale
Single LLM call · no agentic loop
1 agent + deep research mode (35B)
1 agent + deep research mode (397B)
An OS with agent team: Up to 150 agents · 15,000+ steps
Use Case
Synthesizing known answers
Fast, source-grounded answers
Complex technical problems
Open-ended scientific discovery
State-of-the-Art Benchmarks
Apodex Deep Discover results on publicly available deep-research benchmarks.
90.3
SOTA
Multi-hop deep web retrieval benchmark
46.7
FrontierScience Research
SOTA
Open-ended scientific discovery benchmark
94.4
DeepSearchQA
SOTA
Multi-hop deep search question answering
BrowseComp
Apodex Deep Discover
SOTA
90.3
OpenAI GPT-5.5-Pro
90.1
Kimi-K2.6
86.3
Gemini-3.1-Pro
85.9
Claude-Opus-4.8
84.3