Models
The AI models we use, and what each is best at
AI is included in your plan. Transcript Audio picks a sensible default for each job, and Premium users can choose premium models. Here is how the options compare, with honest trade-offs. Updated 30 October 2026.
This is a sample
The full catalogue is maintained alongside the desktop app and is updated as providers change their models. Costs shown are approximate provider costs in US dollars per hour of meeting, to explain the trade-offs. They are included in your plan; you do not pay them separately.
Transcription
Turns the meeting audio into text, ideally with who said what.
Rank 1: ElevenLabs Scribe
- Best quality
Very accurate, with strong speaker separation and accents. One of the pricier options.
Approximate cost: About $0.40 an hour
Rank 2: AssemblyAI Universal
- Recommended
Excellent accuracy and speaker labels at a fair price. A safe default for most meetings.
Approximate cost: About $0.15 an hour
Rank 3: Deepgram Nova-3
- Fastest
Very fast with good accuracy. Slightly weaker than the top two on heavy crosstalk.
Approximate cost: About $0.26 an hour
Rank 4: OpenAI gpt-4o-transcribe
Good accuracy and handy if you already have an OpenAI key. No built-in speaker labels.
Approximate cost: About $0.36 an hour
Rank 5: Google Gemini Flash
Cheap and capable, with a generous free tier on some accounts. Timestamps are less precise.
Approximate cost: About $0.10 an hour
Rank 6: Groq Whisper large-v3-turbo
- Cheapest
About 10x cheaper and very fast, but weaker on accents and people talking over each other.
Approximate cost: About $0.04 an hour
Rank 7: On-device (whisper.cpp / Parakeet)
- Private (on your device)
Free and fully private: audio never leaves your computer. Slower on older machines and less accurate than the best cloud models.
Approximate cost: Free (your hardware)
Notes and summaries
Writes the summary, decisions and action items from the transcript.
Rank 1: Anthropic Claude Opus
- Best quality
The most thorough, well-structured notes and action items. Costs the most per meeting.
Approximate cost: About $0.30 an hour
Rank 2: OpenAI GPT-5 class
Excellent summaries and reasoning at a moderate price. Can be wordy without a tight template.
Approximate cost: About $0.05 an hour
Rank 3: Google Gemini Pro
Strong on long meetings thanks to a very large context window. Formatting is sometimes looser.
Approximate cost: About $0.04 an hour
Rank 4: Anthropic Claude Sonnet
- Recommended
Close to the best quality at a fraction of the cost. The sweet spot for everyday notes.
Approximate cost: About $0.06 an hour
Rank 5: Mistral Large
Good notes from an EU-based provider. Slightly less precise on action items.
Approximate cost: About $0.03 an hour
Rank 6: Google Gemini Flash (or Claude Haiku)
- Cheapest
- Fastest
Pennies per meeting and very fast. Fine for short calls, thinner on nuance in long ones.
Approximate cost: Under $0.01 an hour
Rank 7: Ollama / LM Studio (local model)
- Private (on your device)
Free and private on your own hardware. Quality depends on the model and needs a capable machine.
Approximate cost: Free (your hardware)
Structured fields
Fills template fields such as next steps, owners, budget or dates.
Rank 1: Anthropic Claude Sonnet
- Best quality
- Recommended
Reliable at filling structured fields (budget, next steps, owner) without making things up.
Approximate cost: About $0.03 an hour
Rank 2: OpenAI GPT-5 mini class
Accurate structured output for a low price.
Approximate cost: About $0.01 an hour
Rank 3: Google Gemini Flash
- Cheapest
- Fastest
Very cheap and quick. Occasionally misses fields that are only implied.
Approximate cost: Under $0.01 an hour
Rank 4: Ollama / LM Studio (local model)
- Private (on your device)
Private, but smaller local models fill fields less consistently.
Approximate cost: Free (your hardware)
Meeting titles
Names each meeting so it is easy to find later.
Rank 1: Google Gemini Flash
- Recommended
- Cheapest
A good meeting title costs a tiny fraction of a penny.
Approximate cost: Under $0.01 an hour
Rank 2: Ollama / LM Studio (local model)
- Private (on your device)
Even small local models write decent titles.
Approximate cost: Free (your hardware)
Questions as you go and help
Answers questions about the meeting while it is happening, and in-app help.
Rank 1: Anthropic Claude Haiku
- Recommended
- Fastest
Quick, cheap answers for Questions as you go and in-app help.
Approximate cost: About $0.02 an hour
Rank 2: Ollama / LM Studio (local model)
- Private (on your device)
Private answers, but slower and less accurate on a laptop.
Approximate cost: Free (your hardware)
Presets
Presets pick a model for every job in one click. Which presets and premium models you can choose depends on your plan.
Recommended
Great quality at a sensible price for everyday meetings.
- Transcription
- AssemblyAI Universal
- Notes and summaries
- Anthropic Claude Sonnet
- Structured fields
- Anthropic Claude Sonnet
- Meeting titles
- Google Gemini Flash
- Questions as you go and help
- Anthropic Claude Haiku
Most accurate
The best transcripts and notes, whatever the cost.
- Transcription
- ElevenLabs Scribe
- Notes and summaries
- Anthropic Claude Opus
- Structured fields
- Anthropic Claude Sonnet
- Meeting titles
- Google Gemini Flash
- Questions as you go and help
- Anthropic Claude Haiku
Cheapest
Pennies per meeting with cloud models.
- Transcription
- Groq Whisper large-v3-turbo
- Notes and summaries
- Google Gemini Flash (or Claude Haiku)
- Structured fields
- Google Gemini Flash
- Meeting titles
- Google Gemini Flash
- Questions as you go and help
- Anthropic Claude Haiku
Fastest
Notes ready moments after the call ends.
- Transcription
- Deepgram Nova-3
- Notes and summaries
- Google Gemini Flash (or Claude Haiku)
- Structured fields
- Google Gemini Flash
- Meeting titles
- Google Gemini Flash
- Questions as you go and help
- Anthropic Claude Haiku
Most private
Nothing leaves your computer. Needs a reasonably powerful machine.
- Transcription
- On-device (whisper.cpp / Parakeet)
- Notes and summaries
- Ollama / LM Studio (local model)
- Structured fields
- Ollama / LM Studio (local model)
- Meeting titles
- Ollama / LM Studio (local model)
- Questions as you go and help
- Ollama / LM Studio (local model)
Free
Everything runs on your own hardware, with no cloud processing.
- Transcription
- On-device (whisper.cpp / Parakeet)
- Notes and summaries
- Ollama / LM Studio (local model)
- Structured fields
- Ollama / LM Studio (local model)
- Meeting titles
- Ollama / LM Studio (local model)
- Questions as you go and help
- Ollama / LM Studio (local model)
Prefer to use your own AI provider? Bring your own key is coming soon.