PRODUCT SUITE
Fast, source-grounded research across the full depth of available knowledge. It finds, compares, and synthesizes information into clear intelligence you can act on.
Single-Agent ReAct Architecture
A 35B parameter model running a tight search-synthesize loop. Deep Research browses, extracts, and compiles evidence from across the web into a structured, source-grounded report.
Apodex Deep Research Model
35B parameter model optimized for fast, accurate retrieval and synthesis across dense web sources.
Deep Research Mode
The agent plans a search, executes it, reads the result, and decides the next step — until the answer is complete and generates a report
Where Apodex Deep Research Fits
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
Built for Speed and Accuracy
Apodex Deep Research (35B) on publicly available benchmarks — competitive with leading closed-source systems.
49.8
Hard Deep Research
#1 among all tested systems — beats Claude, GPT-5.4, and Gemini
82.2
Outperforms Gemini-3.1-Pro (81.9) on multi-hop deep search
77.2
FrontierScience-Olympiad
Outperforms OpenAI GPT-5.2 (75.0) on scientific reasoning
Hard Deep Research — vs. Leading Systems
Apodex Deep Research
49.8
Claude Opus 4.7
44.0
OpenAI GPT-5.4
41.3
Gemini 3.1 Pro
40.8
41 hand-authored hard queries. Competitors tested in their respective deep-research modes (Claude Research, GPT-5.4 Deep Research Pro, Gemini Deep Research).