O!TL;DR supports three AI providers for meeting summarization: Anthropic Claude, OpenAI GPT-4, and Google Gemini. Through our BYOK (Bring Your Own Key) feature, you can use any of them with your own API key.
We summarize a lot of meetings. Here is what we have observed.
What makes a good meeting summary?
Before comparing models, it helps to define what good looks like. For recurring meetings specifically, a strong AI summary should:
- Capture decisions clearly, not just topics discussed
- Extract action items with assigned owners and implied deadlines
- Notice what changed compared to previous sessions
- Be concise enough to actually read in 2 minutes
- Handle domain-specific language without confusion
Claude (Anthropic)
Claude is consistently the strongest performer for meeting summarization. It follows complex instructions well, which matters when you have a custom summary template with specific sections and formatting requirements.
Claude is particularly good at:
- Identifying what was actually decided versus what was discussed
- Following nuanced prompt instructions (custom templates work extremely well)
- Structured output -- action items, decisions, and open questions stay cleanly separated
- Handling ambiguity in transcripts (crosstalk, incomplete sentences, filler words)
Best for: Teams with custom summary templates, client-facing summaries that need to be polished, and situations where accuracy of decisions matters most.
Recommended model: Claude Sonnet (cost-effective) or Claude Opus (highest quality).
GPT-4 (OpenAI)
GPT-4 produces reliable, consistent summaries. It is slightly less precise on complex nested decisions, but its outputs are clean and professional.
GPT-4 is particularly good at:
- Producing readable prose summaries that feel natural
- Handling long transcripts without losing coherence
- Familiarity -- teams already using the OpenAI ecosystem find it easy to set up BYOK
- Speed: GPT-4o is notably faster than other frontier models
Best for: Teams already on OpenAI API credits, or situations where summary speed matters more than maximum precision.
Recommended model: GPT-4o (best balance of speed, quality, and cost).
Gemini (Google)
Gemini is the system default for O!TL;DR -- it is what powers summaries for users who have not set up BYOK. It provides strong quality at a lower cost point than Claude or GPT-4.
Gemini is particularly good at:
- Cost efficiency: best output per dollar among the three
- Multilingual handling: especially strong for Korean, Japanese, and Chinese
- Long context: Gemini handles very long transcripts without truncation concerns
Best for: Teams with large meeting volumes watching API costs, multilingual teams, or users who want to try O!TL;DR without setting up a BYOK key first.
Recommended model: Gemini 2.5 Flash (default) or Gemini 2.5 Pro (higher quality).
Summary comparison
| | Claude | GPT-4 | Gemini | |---|---|---|---| | Decision accuracy | Excellent | Good | Good | | Custom template adherence | Excellent | Good | Good | | Multilingual | Good | Good | Excellent | | Speed | Moderate | Fast | Fast | | Cost | Higher | Moderate | Lower | | Setup complexity | Simple | Simple | None (default) |
How to choose
Start with Gemini if you are new to O!TL;DR. It is the default, no setup required, and produces quality summaries.
Switch to Claude if you are using custom summary templates, your meetings involve complex decisions, or you want the highest accuracy on action item extraction.
Use GPT-4 if your team already has OpenAI API credits, or if you are running high-volume summaries and speed is a priority.
Setting up BYOK
In O!TL;DR, go to Workspace Settings, select the AI tab, and paste your API key. It takes about 60 seconds. Once set, all new summaries use your key with no per-summary limits.
Your API key is stored encrypted. We never use your key for anything other than generating your summaries.
Try it free at otldr.com. The first 3 summaries per series are on us.