NextTokenAI/NextSearch-1-S

TEXT GENERATIONPricing:Input $0.4 / Cached $0.07 / Output $4Concurrent Unit Cost:3Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 11, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

NextTokenAI/NextSearch-1-S is a 35.1 billion parameter Mixture-of-Experts (MoE) web research agent developed by NextTokenAI, based on Qwen3.6-35B-A3B. This post-trained model is designed to decompose questions, search and fetch information from the live web, reconcile conflicting evidence, and provide concise answers or structured research artifacts. It is optimized for integration as a research component within larger systems, balancing inference economics with high accuracy on breadth benchmarks. NextSearch-1-S excels in tasks requiring multi-constraint retrieval and comprehensive answer sets from live web data.

Loading preview...

NextSearch-1-S: A Specialized Web Research Agent

NextSearch-1-S is a 35.1 billion parameter Mixture-of-Experts (MoE) model from NextTokenAI, specifically engineered as a web research agent. Built upon the Qwen3.6-35B-A3B base, this model is designed to act as a core research component within larger, orchestrated systems. Its primary function involves decomposing complex questions, performing live web searches, fetching relevant information, and synthesizing findings to deliver concise answers or structured research outputs.

Key Capabilities

  • Intelligent Web Research: Decomposes questions, searches the live web, and fetches information.
  • Evidence Reconciliation: Capable of reconciling conflicting evidence found online.
  • Structured Output: Provides concise answers or structured research artifacts.
  • Optimized for Integration: Designed for repeated calls within larger systems, balancing per-call accuracy, tail latency, and cost.
  • Strong Benchmark Performance: Achieves competitive results on live-web evaluation benchmarks such as FRAMES (0.830), WideSearch-sub (0.737), and WideSearch (0.730), often outperforming other open models in its class.

Ideal Use Cases

NextSearch-1-S is particularly well-suited for applications requiring:

  • Automated Information Retrieval: Systems needing to gather and synthesize information from the live internet.
  • Research Automation: Tools that automate parts of the research process, from query to structured output.
  • Agentic Workflows: As a critical component in multi-agent systems where accurate and cost-effective web interaction is essential.
  • Fact-Checking and Data Synthesis: Tasks involving the reconciliation of diverse online sources to form a coherent answer.

Finetunes
1 models