Integration guides · 2026-09-30
Financial analysis with reasoning and forced tool calls on Ling 3.0 Flash Fin
How to combine the reasoning_effort thinking step and tool_choice required on the finance-focused Ling 3.0 Flash Fin model to build a multi-step analysis flow.
The Fin version is the finance-tuned build of the general Flash
Ling 3.0 Flash Fin is the finance-focused build of the same Ling 3.0 Flash family: tuned for multi-step investment research and long-horizon planning, it carries a 256K context window and has reasoning on by default. Its free offer ended on 30 September 2026 and the row was retired: the identifier now returns 410 model_retired and recommends the metered base model inclusionai/ling-3.0-flash, and requests are never silently rerouted. The thinking and tool-calling flow below works the same way on that model; the differences are that it has no finance tuning and is billed per token.
This article covers how to use the thinking control and tool-calling capabilities listed separately on the model's catalog page together in a financial analysis flow.
When to keep thinking on, when to turn it off
Thinking is on by default, and keeping that default on a multi-step analysis task (comparing multiple data points and drawing a conclusion, for example) lets the model lay out intermediate steps visibly. On a simple query where you just want to pull one piece of data directly from a tool and return it as-is, on the other hand, sending `reasoning_effort: "none"` (or using the `:none` suffix) removes unnecessary thinking time; this is the only level that has a real effect on this model.
At a point where you need a tool call to happen for certain (pulling a current figure from a financial data source, for example), sending `tool_choice: "required"` forces the model to go to the real source instead of guessing the value; this works regardless of the thinking state.
- Keep the default thinking on for multi-step analysis.
- Reduce latency with `reasoning_effort: "none"` on simple, single-data-pull queries.
- Send `tool_choice: "required"` where you need the tool called for certain; this works whether thinking is on or off.
The model accepts text input only
Flash Fin does not accept image or audio input and does not offer schema-enforced JSON output; if you need to send a table or chart from a financial report directly as an image, this model isn't the right fit — you need text-converted data or a different vision-capable model.
Frequently asked questions
Is Ling 3.0 Flash Fin still free?
No. The free offer ended on 30 September 2026 and the row was retired; requests return 410 and recommend the metered inclusionai/ling-3.0-flash. Nothing is charged to your account until you change the model id yourself.
Can I send low or medium for reasoning_effort?
Only the none value produces a real effect on this model; check the current information on the model page before sending other levels.