In collaboration with @nvidia we're releasing two new open weight models: Fastino-Nemotron-3.5-Lightning-Finance and Fastino-Nemotron-3.5-Lightning-Healthcare.
Working closely with the Nemotron team, we developed both models on Nemotron 3.5 Lightning using the Fastino Fine-Tuning Agent, which autonomously ran the entire post-training pipeline. The base model was highly responsive, with the agent reaching substantial gains across both domains with compact datasets alone.
- Fastino-Nemotron-3.5-Lightning-Finance excels at numerical reasoning, research, and summarization over financial concepts and documents. When evaluated, the fine-tuned model increased FinQA execution accuracy by 43.37 points (15.86% to 59.23%), and increased 7.81 points (from 49.65% to 57.46%) on BizFinBench.
- Fastino-Nemotron-3.5-Lightning-Healthcare specializes in clinical conversation quality, summarizing notes and discharge documentation, and extracting medical concepts from unstructured text. In post-training, the Fastino Fine-Tuning Agent improved performance by a margin of 4.28 points on HealthAdminBench (from 25.67% to 29.95%) and 5.09 points on MedCalc-Bench (from 49.09% to 54.18%). It also increased flag accuracy on MEDEC by 11.32 points, beating Opus 4.6 and beating Muse Spark by 0.98 points.
Both models are available on @huggingface under the Apache 2.0 license:
- https://t.co/wJAtFIPAJ5
- https://t.co/6ExYjy3clB
Today we are also releasing the Fastino Fine-Tuning Agent in private preview, which you can sign up for access to here: https://t.co/FtaMnCoCCT
Fine-tuning a small decision model can reduce its classification errors by >50%.
In a recent study, AI One experimented with training a 9B open weight decision model, customized to handle 50 different enterprise workflows.
Using roughly 5,000 synthetic training examples, they increased the model's accuracy from 92.55% to 96.56%, cutting classification errors by more than half and reaching the performance range of the strongest frontier configuration they tested.
Training decision models on domain-specific data is something we've been working to make easier at @fastinoAI.
Our family of open weight decision models, GLiDE (0.4B-27B), is built for fast, accurate decision-making, ranking #1 on the Hugging Face Decision Index.
With the Fastino Training Agent, you can customize GLiDE for your own workflows, from research and synthetic data generation to training and evaluation, without manual tuning.
Send me a message if you're interested in fine-tuning your own open weight models.
Many applications in production depend on making the right call at the right time.
Decision models like GLiDE return fast, structured outputs, making them a great fit for workflows where speed, consistency, and cost matter.
We put together three recipes showing how to use GLiDE for common production tasks:
- Support ticket triage: Classifying and routing incoming requests.
- Agent task gating: Checking proposed agent actions against predefined policies.
- Prompt injection detection: Identifying potentially malicious inputs before they reach your agent.
GLiDE is also available for inference through the Fastino API.
Cookbook: https://t.co/6Ku2drv3va
API: https://t.co/sy82U11REr
got our glide 0.4B at @fastinoAI running on a cpu box with no gpu and no internet :D
it reads a support ticket and answers 3 questions about it in about 90 ms, without generating a single token. what would u run if u didnt need a gpu?
@Airbnb matched frontier-model classification quality at roughly 1% of the serving cost.
By fine-tuning Gemma4-E2B on domain-specific data, Airbnb improved its classifierโs F1 score by roughly 60% while maintaining 35ms median latency. The self hosted model serves thousands of requests per minute.
At @fastinoAI, we've believed in the value of small, customized models from the beginning, and Airbnb's latest engineering case study is proof this approach is gaining traction.
For tasks like guardrails and support triage, teams need models that understand their policies, recognize domain-specific edge cases, and stay reliable at scale.
The results from this case study show how much a small model can gain from training on proprietary data.
That's also why we've been working to make fine-tuning small, open weight models more accessible through our training agent, which handles the process end-to-end with minimal manual intervention.
Weโve opened up our training agent to a select number of enterprises. Reach out if youโre interested in testing.
For enterprise workloads where speed and accuracy matter, smaller, fine-tuned models offer faster, more reliable results than frontier LLMs.
In our latest video, our founding researcher, @urchadeDS shares with @javibuildz, our head of engineering, what motivated him to create the small open weight model, GLiNER, now past 55M downloads on @huggingface, and why he believes smaller models are well-suited for business applications.
A few of the reasons:
- Reliability: Lower risk of hallucinations and out-of-scope outputs
- Privacy: Keeps sensitive data local instead of sending it to external APIs
- Deployment: Easier to run with limited compute resources
- Fine-tuning: More flexibility to customize models for specific tasks
Watch the full conversation: https://t.co/jwTrknWZsH
Download GLiNER: https://t.co/GBqlNXrKll
I fine-tuned 2 GLiDE models for a few dollars on previous festival line up posters and my Spotify listening history - I can now drag and drop new line ups as they drop and see if it fits with my taste and quickly generate playlists.
Try our training API today: https://t.co/JvQa4JUz8a
Technical Thread below
Improving model accuracy doesn't always require additional training.
We increased GLiDE's Decision Index score by 1.85 points using its existing weights.
Common approaches to improving model performance include scaling parameter count, adding training data, or increasing inference-time compute through generated reasoning tokens, but these can all come with significant costs.
Another way is recurrent looping, which increases a model's internal computation by running certain layers multiple times before producing an output.
Some middle layers of transformers operate in a relatively shared representation space, and repeating them can potentially refine the model's internal representations without introducing new parameters or increasing model size.
Our initial experiments focused on training GLiDE to process information through recurrent loops, but these methods added training costs and sometimes even degraded model performance.
Instead, we found that we could apply recurrent loops directly to the already-trained model. By selectively repeating two groups of middle layers, with partial updates to limit changes to the hidden states, GLiDE was able to score 73.37 on the @huggingface Decision Index, surpassing our previously top-ranked model.
We break down the details of our experiments, the results, and key takeaways in our latest blog.
Blog: https://t.co/PSvpcfiUPU
When it comes to AI privacy and security, smaller models may be better suited for the task.
Every interaction with a model introduces potential security risks. Guardrail models are able to detect these risks by classifying inputs and outputs for safety violations, such as harmful content and prompt injection attempts.
Our founding researcher, @urchadeDS, explains how small, customized encoders like GLiGuard can perform these checks quickly, without the overhead of generating responses or reasoning tokens.
Watch the full conversation: https://t.co/Hu8tACYAdc
Download GLiGuard: https://t.co/pWD0UGhgsH