Ling-3.0 Family on Novita AI: Full Lineup Comparison and Which Variant to Use

Start with the request, not the model name. Use Ling-3.0-flash as the general text and agent baseline, Ling-3.0-flash-fin for finance workflows, Ling-3.0-flash-sante for health and medicine, and Ling-3.0-flash-VL when the request includes images or video. All four are serverless chat endpoints on Novita AI with the same 262,144-token context and 32,768-token maximum output, but the differences in domain tuning, input modalities, and current pricing make the model ID a routing decision.

The four models share the same 124B-parameter Flash base. They activate about 5.1B parameters per token, except Flash-VL at about 5.5B. Domain and capability variants are therefore not a ladder from small to large. They are routes through the same architecture.

Key Takeaways

• Text and tool-use baseline: ling-3.0-flash fits general agentic and long-context text workloads.
• Finance tuning: ling-3.0-flash-fin is the option to test for investment, market, and other finance-specific workflows.
• Medical tuning: ling-3.0-flash-sante is the option to test for health and medical reasoning workflows.
• Multimodal route: ling-3.0-flash-vl is the only family member on Novita AI that currently accepts image and video input alongside text.

Ling-3.0 Lineup at a Glance

The model catalog was checked on September 9, 2026. Pricing and availability can change, so recheck the live model page before you hard-code an estimate or production route.

Model ID	Best fit	Total / active parameters	Modalities	Context / max output	Current price per 1M tokens
inclusionai/ling-3.0-flash	General agentic inference, reasoning, and tool use	124B / ~5.1B	Text → text	262,144 / 32,768	$0.06 input; $0.18 output
inclusionai/ling-3.0-flash-fin	Finance-specific reasoning and multi-step workflows	124B / ~5.1B	Text → text	262,144 / 32,768	$0 input; $0 output
inclusionai/ling-3.0-flash-sante	Health and medical reasoning plus general capability	124B / ~5.1B	Text → text	262,144 / 32,768	$0 input; $0 output
inclusionai/ling-3.0-flash-vl	Screenshots, charts, documents, video, and visual agents	124B / ~5.5B	Text + image + video → text	262,144 / 32,768	$0 input; $0 output

The catalog also lists serverless access, reasoning, and function calling for every model. ling-3.0-flash-vl and ling-3.0-flash-sante additionally expose an Anthropic-compatible endpoint; the other two list chat/completions.

What the Variants Share

Every model is built on Ling-3.0-flash. Novita’s catalog and the official Flash model card describe the base Flash route as a 124B-parameter MoE with about 5.1B active parameters per token; Flash-VL is also 124B but activates about 5.5B. That means the choice among the Flash variants is mainly about domain emphasis, input modality, pricing state, and latency behavior, not a difference in total architecture scale.

All four listings currently show the same Novita context and output ceilings: 262,144 tokens in and 32,768 tokens out. That equality makes context a poor first filter. Fit, modality, cost, and measured latency are better ways to narrow the lineup before you benchmark.

Choose Ling-3.0-flash for the General Baseline

Use Ling-3.0-flash when your workload is text-only and does not yet have a clear vertical specialization. It is the natural first test for chat, document analysis, multi-step agents, reasoning controls, and tool calls. The quick start shows the OpenAI-compatible request flow.

It is also the fair control model for comparisons. If Fin or Sante does not improve your task-specific traces, you have evidence for keeping the general route instead of adding an extra model ID.

Choose Ling-3.0-flash-fin for Finance Workflows

Use Ling-3.0-flash-fin when the task depends on finance vocabulary and multi-step financial reasoning. Novita’s listing describes the model as finance-enhanced and suited to real-world investment workflows, long-horizon planning, and execution while retaining reasoning, coding, and math capability.

It is not a compliance system or a substitute for your data controls. Keep citations, source documents, calculations, and approval steps in the application. For portfolio analysis, earnings review, research synthesis, or investment-process drafting, compare Fin with the general Flash model on examples from your actual domain.

Choose Ling-3.0-flash-sante for Health and Medical Tasks

Use Ling-3.0-flash-sante when the request involves medical knowledge reasoning, clinical safety language, evidence retrieval, or long-horizon medical workflows. Novita describes Sante as built on Ling-3.0-flash, with 124B total parameters and about 5.1B active parameters. A chat completions quick start is also available.

Medical use adds requirements beyond model choice: privacy handling, human review, escalation paths, provenance, and local regulatory rules. Sante is the family member to evaluate for medical phrasing and reasoning, but your product should still verify outputs before any clinically consequential action.

Choose Ling-3.0-flash-vl for Image and Video Requests

Use Ling-3.0-flash-VL when the model must see something: screenshots, charts, forms, documents, product imagery, or short videos. It is the only variant in the current Novita listing that accepts image and video input with text output.

The useful test is whether visual evidence changes the action the system takes. A browser agent might read a screen, choose a control, call a tool, and verify the result; a reporting assistant might extract a series from a chart. For text-only prompts, send the request to the base Flash model unless your tests show a reason to use VL.

A Practical Routing Rule

Use a small configuration instead of scattering model IDs through your code:

model_routes = {
    "general_text_agent": "inclusionai/ling-3.0-flash",
    "finance_workflow": "inclusionai/ling-3.0-flash-fin",
    "medical_workflow": "inclusionai/ling-3.0-flash-sante",
    "image_or_video": "inclusionai/ling-3.0-flash-vl",
}

Then let workload metadata choose the route. Start with general Flash for text-only agent work and VL whenever image or video content is present. Add Fin or Sante only when your domain traces show that the specialized route is worth the extra evaluation and operations overhead.

Price the Decision, Not the Name

Current catalog prices split into two patterns. The general Flash model lists $0.06 per million input tokens and $0.18 per million output tokens. Fin, Sante, and VL list $0 for both. Those zero-price listings are live promotion snapshots, not permanent guarantees.

For real production estimates, record prompt size, response size, cache behavior, retries, tool-call frequency, and the number of model turns per user action. A route with a higher list price can be cheaper if it resolves a task in fewer turns or less rework; a free listing does not remove the cost of tests, operations, or quality review.

FAQ

How many Ling-3.0 models does Novita AI offer?

The checked catalog lists four: ling-3.0-flash, ling-3.0-flash-fin, ling-3.0-flash-sante, and ling-3.0-flash-vl.

Do all four models have the same context window?

Yes. Novita’s current listings show 262,144 tokens of context and 32,768 tokens of maximum output for every model.

Which Ling-3.0 model should I use for screenshots or video?

Use inclusionai/ling-3.0-flash-vl. It is the family member that Novita currently lists with text, image, and video input.

Are the free Ling-3.0 prices permanent?

No. Fin, Sante, and VL currently list $0 input and $0 output, but pricing is a live catalog value. Recheck it before production use.

Recommended Articles

• Ling-3.0-flash on Novita AI: Free API for Agentic Inference
• Ling-3.0-flash API Quick Start for Function Calling and Reasoning in Agent Workflows
• Ling-3.0-Flash-VL on Novita AI: Native Multimodal API and Pricing