Need help? You're in the right place.
Cohesion aims to be the best scratchpad for your thoughts. While having a physical scratchpad is awesome for smaller ideas (I love my Field Notes!), sometimes ideas are best expressed through speaking. Cohesion captures those spoken sparks instantly, turning them into searchable, structured notes you can revisit later.
Instead of juggling between voice memos, half-written notes, or trying to type while you’re in the flow, Cohesion gives you one simple place where your ideas can live, grow, and connect. Think of it as a bridge between the spontaneity of speaking and the clarity of writing—a scratchpad designed for your voice.
We strive to achieve a balance between freedom of model selection while still not making the choice overwhelming. That being said, here are results from our testing:
| Model Name | Provider | Availability | Notes |
|---|---|---|---|
| Foundation Model | Apple | Free | This model runs completely on your device. While it may not always have the strongest results and can only support smaller notes, the privacy guarantees of never leaving your device may be appealing. |
| GPT-5 nano | OpenAI | Basic | A great model that's very fast. It may not be as verbose or extract as much text, but the speed may be worth it for you. |
| GPT-5 mini | OpenAI | Pro | This model is slower than nano but can often produce better results. Results are consistently better than on-device models. |
| GPT-5 (regular) | OpenAI | Pro | This model is definitely one of the slower one, but it is extremly good at making consistently good outputs. It follows rules well and will make great results almost all of the time. |
| Gemini 2.5 Flash Lite | Basic | This model is extremely fast. Use this for smaller notes where speed is more important than verbosity. | |
| Gemini 2.5 Flash | Basic | A moderately fast and verbose model. This model produces very good results for it's "weight class". | |
| Gemini 2.5 Pro | Pro | This model isn't as fast as it's Flash bretheren, but it produces better output. We have found the quality to be just below GPT-5 mini's results while being faster. | |
| GPT-OSS | Fireworks | Basic | This model is extremly fast. It provides a quality almost as good as GPT-5 mini and is only slightly slower than GPT-5 nano. |
| Claude Haiku 3 | Anthropic | Basic | A very fast model with human-feeling results. Anthropic has also historically been adamant about privacy. |
| Claude Haiku 3.5 | Anthropic | Pro | Slightly slower than Haiku 3 but with more consistent results. Anthropic has also historically been adamant about privacy. |
| Claude Sonnet 4 | Anthropic | Pro | Anthropic's flagship model. This does cost a lot so you won't get as many requests with it, but the results are high quality. Our recommendation is to use this for the most complex notes. |
Note that just because we see these results it doesn't mean that you will. Try out different models to find what works best for you and how you speak.
Cohesion goes the extra mile to keep your notes as your notes. From offering on-device models on supported devices, to ensuring our providers are up to our standard, we make sure your data is safe. For more info, check out our human-readable privacy policy.
Most of Cohesion's features are actually entirely free. Transcription and on-device models are included for free with the app (on supported devices). If you decide you want to level up your note-taking and use more powerful models, you can subscribe to our service to get access to cloud models. At the time of writing, there are two tiers with different levels of access. With the basic tier, you get a limited set of models with a handful of requests per month for $2.99/month USD. With the pro tier, you can access all models we enable with way more requests per month for $6.99/month. These numbers may be out of date, check out the settings page for the most up-to-date pricing.
Most of the models in Cohesion require us to pay a fee to the model provider depending on the length of your recording (this is called per-token pricing). We aim to reduce the confusion of per-token pricing by offering a flat-rate subscription. Additionally, some of the subscription fees go into making the app better.