InclusionAI released Ling 3.0 Flash on July 23, 2026. It is a 124B total parameter Mixture-of-Experts model with approximately 5.1B parameters activated per token. The key innovation is a hybrid reasoning mode that combines the speed of the Ling series with the deep reasoning of the Ring series. It is free on OpenRouter through August 3, 2026.
This is the successor to Ling 2.6 Flash, which had 104B total parameters and 7.4B active. Ling 3.0 Flash has more total parameters (124B vs 104B) but fewer active per token (5.1B vs 7.4B), making it more efficient. The context window also doubled from 128K to 262K.
Here is everything about Ling 3.0 Flash: specs, benchmarks, how it compares to other Flash models, and how to use it.
Key Specs
| Spec | Value |
|---|---|
| Release date | July 23, 2026 |
| Total parameters | 124B |
| Active parameters | ~5.1B per token |
| Architecture | Mixture-of-Experts (MoE) |
| Context window | 262K (extendable to 1M) |
| Reasoning mode | Hybrid (Ling speed + Ring reasoning) |
| Open weights | Expected (not yet released) |
| API pricing | Free on OpenRouter (limited time) |
| Developer | InclusionAI (Ant Group) |
The 124B/5.1B split is the key number. Total parameters determine what the model knows. Active parameters determine what it costs to run. At 5.1B active, Ling 3.0 Flash is in the same inference class as Gemini 3.5 Flash-Lite but with significantly more total knowledge.
What Changed from Ling 2.6 Flash
Ling 3.0 Flash builds on the Ling 2.6 family with three key improvements:
More total parameters, fewer active. 124B total (up from 104B) with 5.1B active (down from 7.4B). This means more knowledge capacity with lower inference cost.
Hybrid reasoning mode. The Ling series was designed for speed. The Ring series was designed for reasoning. Ling 3.0 Flash combines both. The model dynamically scales its thinking effort depending on task difficulty, delivering logic precision when needed without wasting tokens on simpler prompts.
Doubled context window. 262K tokens (up from 128K), extendable to 1M. This puts it in the same class as Gemini 3.6 Flash and Claude Sonnet 5 for context length.
How to Use Ling 3.0 Flash
Via OpenRouter (Free)
Ling 3.0 Flash is free on OpenRouter through August 3, 2026:
import openai
client = openai.OpenAI(
base_url="https://openrouter.ai/api/v1",
api_key="your-openrouter-key"
)
response = client.chat.completions.create(
model="inclusionai/ling-3.0-flash",
messages=[
{"role": "user", "content": "Write a Python function that implements binary search."}
]
)
print(response.choices[0].message.content)For more on OpenRouter setup, see our OpenRouter Complete Guide.
With Coding Tools
Ling 3.0 Flash works with any tool that supports OpenAI-compatible endpoints:
- Aider: Set the model to
openrouter/inclusionai/ling-3.0-flash - Continue: Add an OpenRouter provider pointing to Ling 3.0 Flash
- OpenCode: Configure the OpenRouter endpoint as a custom model
Pricing
| Provider | Price | Notes |
|---|---|---|
| OpenRouter | Free | Limited time (through August 3, 2026) |
| API | Not yet announced | Expected to be very affordable |
The free tier is generous enough for evaluation and development. For production use at scale, you will need to watch for the paid pricing announcement.
For a full pricing comparison across models, see our AI API Pricing Compared 2026 guide.
How Ling 3.0 Flash Compares to Other Flash Models
| Spec | Ling 3.0 Flash | Gemini 3.6 Flash | Gemini 3.5 Flash-Lite |
|---|---|---|---|
| Total parameters | 124B | Not disclosed | Not disclosed |
| Active parameters | ~5.1B | Not disclosed | Not disclosed |
| Context window | 262K | 1M | 1M |
| Reasoning mode | Hybrid | Yes | Yes (configurable) |
| Pricing | Free (limited) | $1.50/$7.50 | $0.30/$2.50 |
| Open weights | Expected | No | No |
Ling 3.0 Flash has the advantage of being open-weight (expected) and free (limited time). Gemini 3.6 Flash has a larger context window (1M vs 262K) and built-in computer use. Gemini 3.5 Flash-Lite is cheaper at scale ($0.30/$2.50).
For a full comparison, see our Ling 3.0 Flash vs Gemini 3.6 Flash article.
Local Setup (When Weights Are Released)
Ling 3.0 Flash is expected to be open-weight on HuggingFace. When the weights are released, you will be able to run it locally:
# With Ollama (when available)
ollama run inclusionai/ling-3.0-flash
# With vLLM (when weights are released)
python -m vllm.entrypoints.openai.api_server \
--model inclusionai/ling-3.0-flash \
--dtype autoFor hardware requirements, you will need approximately:
- Minimum: 16 GB VRAM with quantization
- Recommended: 24 GB VRAM for comfortable inference
- Speed: Expect 20-40 tok/s on modern GPUs
See our Ollama Complete Guide 2026 and How Much VRAM for AI for more on local setup.
My Take
Ling 3.0 Flash is the most efficient open-weight model available right now. At 124B total parameters with only 5.1B active, it has the knowledge capacity of a large model with the inference cost of a small one.
The hybrid reasoning mode is the real innovation. Most models force you to choose between speed (Flash models) and reasoning (Pro/Thinking models). Ling 3.0 Flash combines both. For tasks that need reasoning, it activates the Ring-inspired thinking. For simple tasks, it stays fast.
The free tier on OpenRouter is generous enough for evaluation. If you are building agents or coding tools, this is worth testing. The 262K context window is enough for most codebases, and the open weights (when released) will make self-hosting viable.
My one concern: the context window is smaller than Gemini 3.6 Flash (262K vs 1M). For tasks that involve very large codebases or documents, Gemini has the edge. For everything else, Ling 3.0 Flash is competitive.
For comparisons with other models, see our Ling 3.0 Flash vs Ling 2.6 Flash and Ling 3.0 Flash vs DeepSeek V4 articles.
FAQ
Is Ling 3.0 Flash free?
Yes, on OpenRouter through August 3, 2026. After that, paid pricing will apply (not yet announced). The free tier is generous enough for evaluation and development.
How does Ling 3.0 Flash compare to Gemini 3.6 Flash?
Ling 3.0 Flash has more total parameters (124B vs not disclosed) and is open-weight (expected). Gemini 3.6 Flash has a larger context window (1M vs 262K), built-in computer use, and costs $1.50/$7.50. See our Ling 3.0 Flash vs Gemini 3.6 Flash comparison.
Can I run Ling 3.0 Flash locally?
Not yet. The weights are expected to be released on HuggingFace. When they are, you will need approximately 16 GB VRAM with quantization to run it locally.
What is hybrid reasoning?
Ling 3.0 Flash combines the speed of the Ling series with the reasoning capability of the Ring series. The model dynamically scales its thinking effort depending on task difficulty. Simple tasks get fast responses. Complex tasks activate deeper reasoning.
What coding languages does Ling 3.0 Flash support?
Ling 3.0 Flash is trained on code across all major programming languages. It handles Python, JavaScript, TypeScript, Java, C++, Go, Rust, Ruby, PHP, Swift, Kotlin, and many others.