Released 2026-07-23 · general agent
Deployment evidence
BAAI · 4B dense total · 4B dense active · 256K tokens
Best-fit scenarios
Compact deep-research model for long-horizon information seeking, evidence aggregation and multi-constraint verification.
Not recommended
- Treating long-context support as proof of source reliability
- Deploying research answers without citation and retrieval validation
Deep research agents
Vendor-statedBAAI positions AREX-Turbo for long-horizon information seeking, evidence aggregation and multi-step research.
Capability sourceMulti-constraint verification
Public evidenceThe official release publishes deep-research and verification evaluations; results are vendor measurements rather than ChinaAPI reproductions.
Capability sourcePrivate research endpoint
ChinaAPI inferenceThe dense 4B shape and three official inference paths make it a practical private research-agent candidate.
Official launch / minimum
Evidence COfficial paths no hardware floor
Transformers, vLLM and SGLang launch paths are documented, but the vendor does not publish an absolute minimum GPU or memory configuration. Primary recipe
200-person team
Evidence EBenchmark required
Start with one measured worker and size by concurrent research trajectories, tool calls and context accumulation rather than employee count.
Commercial API
Evidence COfficial framework paths
Official vLLM and SGLang servers are available; public use still needs replicas, retrieval observability and citation-quality controls. Primary recipe
Commercial-use check
Apache-2.0
MaaS / hosted service: No model-specific MaaS restriction identified in Apache-2.0.
Attribution: Provide the license and required notices, preserve attribution notices, and mark modified files; trademark rights are not granted.
Known limitations and open questions
- No ChinaAPI reproduction or official hardware minimum
- Vendor evaluations do not validate a specific retrieval stack
- Long-horizon quality depends on tools, sources and orchestration outside the weights