Meet Fin Apex 2.0 The best-performing model
for customer service

Introducing the
Fin model suite

• [ 1 ]

  Fin Apex 2.0 [ new ]

  

  Apex 2.0 is the generative core of Fin. It takes the most relevant content retrieved and ranked by the Fin model suite, applies your configured policies, and produces a direct answer or decides the question needs a human. It builds on what we learned from Apex 1.0 in production, combining a stronger base model with new post-training data. The result is 8.6% more confirmed resolutions and higher answer accuracy.

  Specification

  • 1.1Grounded in your knowledge base, not general data
  • 1.2Answers directly, escalates honestly when it can't
  • 1.3Asks for clarification instead of guessing
  • 1.4Built on learnings from Apex 1.0 in production
  • 1.58.6% more confirmed resolutions vs. Apex 1.0
  • 1.616.5% fewer hallucinations vs. Apex 1.0
• [ 2 ]

  Fin Apex Flash 2.0 [ new ]

  Apex Flash 2.0 is the next generation of Apex Flash, our lightweight version of Apex for latency-sensitive channels like Voice. It answers 24% faster and has 46% fewer hallucinations than the previous version.

  Specification

  • 2.1Built for latency-sensitive customer interactions
  • 2.2Grounded in your knowledge base, not general data
  • 2.324% faster answers vs. Apex Flash 1.0
  • 2.4Shorter answers that suit spoken conversation
• [ 3 ]

  Fin Apex for Procedures [ new ]

  Apex for Procedures is the model behind Fin Procedures. It reasons through complex, multi-step queries, follows your business rules at each step, and decides what Fin should do next. It outperforms Sonnet 5, the best-performing frontier model we tested for this work.

  Specification

  • 3.1Reasons through complex, multi-step queries
  • 3.2Follows your rules and guidance at each step
  • 3.315% more confirmed resolutions vs. Sonnet 5
  • 3.437% fewer hallucinations vs. Sonnet 5
• [ 4 ]

  Fin Retrieval

  The Fin Retrieval model scans all available knowledge sources and pulls out a small set of useful information.

  Specification

  • 4.1Understand intent behind the question
  • 4.2Search across all available content sources
  • 4.3Match on semantics, not just keywords
  • 4.4Select top N most likely candidates
• [ 5 ]

  Fin Reranker

  The Fin Reranker model takes the retrieved content and scores each piece for relevance, accuracy, and usefulness in context. It then selects the final piece(s) for the LLM to use.

  Specification

  • 5.1Score relevance with respect to query
  • 5.2Evaluate context match and resolution fit
  • 5.3Downrank outdated or low-confidence sources
  • 5.4Output: final selected content for generation
• [ 6 ]

  Fin Issue Summarizer

  The Fin Issue Summarizer model detects and extracts the user's issue from conversation history. Built on a fine-tuned 14B model, it transforms unstructured chat exchanges into clear, actionable issue summaries for downstream retrieval.

  Specification

  • 6.1Detect whether conversation contains an addressable issue
  • 6.2Extract summarized issue from multi-turn exchanges
  • 6.3Handle edge cases: greetings, feedback, and noise
  • 6.412.5% cost reduction vs frontier LLMs
• [ 7 ]

  Fin Feedback Parser

  The Fin Feedback Parser model uses a multi-task ModernBERT architecture to interpret user responses. It classifies feedback sentiment, detects follow-up questions, and identifies conversation endings with state-of-the-art accuracy.

  Specification

  • 7.1Classify feedback type: positive, negative, or none
  • 7.2Detect follow-up questions in user messages
  • 7.3Identify when conversations have ended
  • 7.4Match LLM accuracy at a fraction of the cost
• [ 8 ]

  Fin Language Detector

  The Fin Language Detector model uses XLM RoBERTa to accurately identify the user's language across 45 supported languages. It handles real-world challenges like typos, short messages, and script mismatches.

  Specification

  • 8.1Support 45 languages with high accuracy
  • 8.2Handle spelling mistakes and short context
  • 8.3Process script mismatches (e.g., Romanized Hindi)
  • 8.4Enable intelligent language fallback logic
• [ 9 ]

  Fin Escalation Router

  The Fin Escalation Router model uses a multi-task ModernBERT architecture to decide when conversations should be handed off to human agents. It provides reasoning and cites matching business guidelines with over 98% accuracy.

  Specification

  • 9.1Decide: escalate, offer to escalate, or continue
  • 9.2Cite matching business escalation guidelines
  • 9.3Provide reasoning for escalation decisions
  • 9.40.5s faster than LLM-based routing