GPT-6 Luna
MiMo-V2.6-Flash
7.9B- input
- $0.042
- cached
- $0.0084
- output
- $0.1232
- Context
- 262Ktokens
- Released
- 2mo ago
- Size
- 127Bparameters
Cost/1M tokens
Finance-enhanced model for financial research, multi-step investment workflows, and long-horizon planning and execution
- Language model
- Open weights · 127B
- Reasoning
- Tools
How it compares
Level
LightweightThe lab's smallest, cheapest and lowest-latency hosted line, built for high volume and simple tasks.
Lightweight models from other labs
Vs. your stack
Add the models you use to see how this one compares.
Sign inFamily tree
Models trained from one another, as their makers describe it.
Ling 3.0 FlashInclusionAI · 2026
Additional source evidence
- Source model
- inclusionai/ling-3.0-flash-fin
- Source family
- ling
- Last updated
- 2026-09-03
- Model context
- 262K
- Model output limit
- 33K
- Open weights
- Yes
- File attachments
- No
- Reasoning
- Yes
- Tool calling
- Yes
- Temperature control
- Yes
Downloadable weights
Published benchmarks
| Benchmark | Score | Evaluation details | Evidence |
|---|---|---|---|
| FinFIRST | 52.85 · score | ReAct with web search, visit, and Python · highest reasoning effort | Source |
| FinSearchComp Verified | 77.04 · score | ReAct with web search, visit, and Python · expert-validated internal set · highest reasoning effort | Source |
| FinCRAFT | 51.74 · score | internal · highest reasoning effort | Source |
| Finance Agent | 69.19 · strict-pass | highest reasoning effort; 10 runs · 1.1 | Source |
| Finance Agent | 59.81 · strict-pass | highest reasoning effort; 20 runs · 2 | Source |
| APEX-Agents | 29.17 · score | highest reasoning effort | Source |
| SpreadsheetBench | 86.5 · score | Claude Code 2.1.173 · highest reasoning effort; temperature=0.6; 120 turns; 3h timeout · 1 | Source |
| SpreadsheetBench | 21.81 · score | Claude Code 2.1.173 · highest reasoning effort; temperature=0.6; 300 turns; 3h timeout · 2 | Source |
| tau3-bench | 41 · score | highest reasoning effort; banking | Source |
Complete source record (JSON)
{
"id": "inclusionai/ling-3.0-flash-fin",
"name": "Ling 3.0 Flash Fin",
"description": "Finance-enhanced model for financial research, multi-step investment workflows, and long-horizon planning and execution",
"family": "ling",
"last_updated": "2026-09-03",
"attachment": false,
"reasoning": true,
"tool_call": true,
"temperature": true,
"modalities": {
"input": [
"text"
],
"output": [
"text"
]
},
"limit": {
"context": 262144,
"output": 32768
},
"benchmarks": [
{
"name": "FinFIRST",
"score": 52.85,
"source": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin",
"metric": "score",
"harness": "ReAct with web search, visit, and Python",
"variant": "highest reasoning effort"
},
{
"name": "FinSearchComp Verified",
"score": 77.04,
"source": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin",
"metric": "score",
"harness": "ReAct with web search, visit, and Python",
"variant": "highest reasoning effort",
"dataset": "expert-validated internal set"
},
{
"name": "FinCRAFT",
"score": 51.74,
"source": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin",
"metric": "score",
"variant": "highest reasoning effort",
"dataset": "internal"
},
{
"name": "Finance Agent",
"score": 69.19,
"source": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin",
"metric": "strict-pass",
"variant": "highest reasoning effort; 10 runs",
"version": "1.1"
},
{
"name": "Finance Agent",
"score": 59.81,
"source": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin",
"metric": "strict-pass",
"variant": "highest reasoning effort; 20 runs",
"version": "2"
},
{
"name": "APEX-Agents",
"score": 29.17,
"source": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin",
"metric": "score",
"variant": "highest reasoning effort"
},
{
"name": "SpreadsheetBench",
"score": 86.5,
"source": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin",
"metric": "score",
"harness": "Claude Code 2.1.173",
"variant": "highest reasoning effort; temperature=0.6; 120 turns; 3h timeout",
"version": "1"
},
{
"name": "SpreadsheetBench",
"score": 21.81,
"source": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin",
"metric": "score",
"harness": "Claude Code 2.1.173",
"variant": "highest reasoning effort; temperature=0.6; 300 turns; 3h timeout",
"version": "2"
},
{
"name": "tau3-bench",
"score": 41,
"source": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin",
"metric": "score",
"variant": "highest reasoning effort; banking"
}
],
"release_date": "2026-08-27",
"open_weights": true,
"license": "MIT",
"weights": [
{
"label": "Hugging Face",
"url": "https://huggingface.co/inclusionAI/Ling-3.0-flash-Fin"
}
]
}Frequently asked questions
Running locally
At Q4 it needs an estimated 89.4 GB, at Q8 156.3 GB and at FP16 293.7 GB. All figures are estimates from the 127B parameter count, weights plus about 15% and 0.5 GB for a short context.
No. By estimate it does not fit on a 24 GB, 16 GB, 12 GB or 8 GB GPU at Q4, Q8 or FP16.
Yes, but only larger Macs. Q4 needs an estimated 89.4 GB of VRAM and macOS is assumed to let the GPU use about 75% of unified memory, so the smallest common Mac memory size that fits is 128 GB (estimate).
API and pricing
$0.042 per 1M input tokens and $0.1232 per 1M output tokens. This comes from 1 tracked paid provider.
Yes. Cached input costs $0.0084 per 1M tokens versus $0.042 uncached, an 80% discount.
$0.0005432 at the cheapest provider, Novita, for a request with 10,000 input tokens and 1,000 output tokens.
1 endpoint had uptime data in the last day and it was at 99% uptime or better. Best uptime was 100.0%, by Novita.
Privacy and data
Novita is the tracked provider with zero data retention.
Yes. Ling 3.0 Flash Fin has open weights, so you can self-host it for full control of your data.
Capabilities and specs
The context window is 262K tokens (262,144) and the maximum output is 33K tokens (32,768).
No, it only accepts text. Input modalities: text.
Yes, tool calling (function calling) is supported.
Yes, a reasoning mode is supported, but reasoning is not always on. 1 of 1 tracked providers exposes it and no provider offers a reasoning_effort setting.
Quality and benchmarks
Nex-N2.5-Mini by Nex AGI at $0.0438 per 1M tokens, GLM 5.3 Flash by Z.ai at $0.0495 and Granite 4.2 8B by IBM at $0.0825 (blended prices). Matches are based on model kind, blended price and release date.
Nex-N2.5-MiniLing 3.0 Flash Fin vs Nex-N2.5-MiniGranite 4.2 8BLing 3.0 Flash Fin vs Granite 4.2 8BGLM 5.3 FlashLing 3.0 Flash Fin vs GLM 5.3 Flash
Versions and lineage
Both have a 262K token (262,144) context window. Ling 3.0 Flash Fin costs $0.042 input and $0.1232 output per 1M tokens and was released August 27, 2026. Ling 3.0 Flash costs $0.021 input and $0.063 output per 1M tokens and was released July 23, 2026.
Ling 3.0 Flash Fin is fine-tuned from Ling 3.0 Flash.
Licensing and access
It is open weights, licensed under MIT.
Yes. The MIT license is permissive and allows commercial use.
From the Hugging Face repository inclusionAI/Ling-3.0-flash-Fin.
Basics
InclusionAI, released August 27, 2026.
No. The listed API price is $0.042 per 1M input tokens and $0.1232 per 1M output tokens, though the model has open weights.
Generated from Vertical Tabs data, updated November 12, 2026.
Discussion
No comments yet
Be the first to share your thoughts.