Open Finetuning

What it does

Openfinetuner, from Lyzr, takes in your documents from sources such as S3 buckets and SharePoint folders and turns them into training data. It produces fine-tuned open-weight models, using LoRA and DPO, that are scored against your own test cases and served from an inference endpoint. A built-in model router then decides, at runtime, whether each task goes to a frontier model, an approved general-purpose model or one of your fine-tuned models.

Teams use it for narrow, high-volume work such as classification, extraction and document question answering, where the same schema and answer shape repeat. A model must clear the evals before it serves traffic, and scheduled evals and refinetuning keep it checked in production. Customers supply and pay for their own GPUs, storage and third-party model APIs.

This suits enterprises with large token bills on repetitive tasks and the GPU capacity or cloud footprint to host models. The listed pricing is an annual license of $250,000 that includes five production model lineages. Tasks such as code generation and heavy multi-step reasoning are kept on frontier models.

What it can do

• Live document sync from S3 and SharePoint with topic sorting and de-duplication
• Generation of QA pairs and preference pairs, with synthetic data where coverage is thin
• LoRA supervised fine-tuning followed by DPO on three candidate base models by default
• Continuous evaluation of candidate models against test cases compared with a frontier model
• Refinetuning loop that retrains when evals fail and ships only models that clear benchmarks
• Built-in model router that selects among frontier, general-purpose and fine-tuned models with a recorded reason per decision

Works with

• AWS
• Azure
• Google Cloud
• S3
• SharePoint
• vLLM

Before you choose

• Frontier models remain the default for code generation, long-form writing and heavy multi-step reasoning, which the product does not target.
• Customers pay separately for GPUs, storage and third-party model APIs; the license does not include them.
• Additional production model lineages beyond five cost $35,000 per year each, and the $250,000 figure is labeled a pricing proposal.

Listed under

• Coding
• MCP and agent tools

Get started

The page directs visitors to send a month of token usage by workload to the Openfinetuner team for a review and savings estimate.

1. Connect a document source such as an S3 bucket or SharePoint folder so Openfinetuner sets up a live sync.
2. Define the test cases that a fine-tuned model must clear against a frontier baseline.
3. Choose base models and training methods per lineage in a declarative config, then run LoRA and DPO and review the eval results before promotion.

Annual license covers the platform. Development experiments are described as unlimited, subject to technical capacity and reasonable-use and support terms, with compute supplied by the customer. Retraining within a lineage does not count as a new lineage.

Source: openfinetuner.lyzragents.com, checked 29 Oct 2026. Prices can change; check before you buy.

Sources for this overview

Product pages, documentation, and pricing checked 29 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Lyzr Open FineTuner | Route work to the model that earned it ↗openfinetuner.lyzragents.com
• Lyzr | Take your AI agents to production, faster. ↗www.lyzr.ai

Company website

openfinetuner.lyzragents.com: product, pricing and policy pages

Web search

Confirms the official site

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

manual_review

Last checked 29 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.