We've upgraded the AI backbone of FastScribeX across two major feature areas. Here's what changed, what we switched and why, and what it means for your day-to-day workflow.
Summary of Changes
| Feature | Previous | Now |
|---|---|---|
| AI Summary | GPT-5.2 + Claude Opus 4.6 (fixed) | GPT-5.4, Claude Opus 4.7, and more - your choice |
AI Summaries: GPT-5.4 and Claude Opus 4.7
What changed
AI summaries in FastScribeX are now powered by OpenAI's GPT-5.4 and Anthropic's Claude Opus 4.7 - the two most capable large language models commercially available in 2026. You can choose which model generates your summary, or let the system select the optimal one based on your content type.
Why these two models
We tested nine models across a library of 400+ real-world transcripts - meetings, interviews, earnings calls, lectures, and podcasts. GPT-5.4 and Claude Opus 4.7 consistently outperformed the others on the criteria that matter most:
| Evaluation Criterion | Why It Matters |
|---|---|
| Long-context retention | Summaries of 2+ hour recordings that miss nothing from the final 30 minutes |
| Instruction following | Summaries that precisely match the requested template (action items, decisions, topics) |
| Speaker attribution accuracy | Correctly attributing summary points to the right speaker in multi-speaker recordings |
| Hallucination rate | Not adding invented content that wasn't in the original audio |
| Technical domain handling | Legal, medical, and financial terminology handled correctly without paraphrasing into errors |
GPT-5.4 vs Claude Opus 4.7 - when to use each
Both models produce high-quality summaries, but they have different strengths:
| GPT-5.4 | Claude Opus 4.7 |
|---|---|
| Faster processing | Slightly slower but often richer output |
| Stronger on structured output (action items, numbered lists) | Stronger on narrative synthesis and nuanced tone |
| Excellent for business meeting formats | Excellent for interview, academic, and editorial content |
| Better at following strict formatting templates | Better at free-form summary quality |
What the summary includes
Every AI summary delivers:
- Key Decisions - What was decided, by whom, and the context
- Action Items - Tasks assigned, deadlines mentioned, owners named
- Main Topics - A structured outline of what was discussed
- Key Quotes - Notable verbatim statements worth preserving
- Open Questions - Items left unresolved at the end of the recording
The summary template is customizable on Pro and Business plans - you can add, remove, or rename sections to match your team's format.
AI Chat: Gemini 3 Flash
What changed
AI Chat - the feature that lets you ask natural-language questions about any transcript and get sourced, timestamped answers - is now powered by Google's Gemini 3 Flash.
Why Gemini 3 Flash
We chose Gemini 3 Flash for AI Chat after extensive benchmarking against other models specifically for retrieval-augmented generation (RAG) tasks - the technical name for what AI Chat does: reading a long document and answering specific questions about it.
The results were clear:
| Benchmark | Gemini 3 Flash Performance |
|---|---|
| Answer relevance | Top-tier for precise, scoped answers |
| Latency | Significantly faster response times than alternatives |
| Multi-turn coherence | Maintains context across follow-up questions accurately |
| Source attribution | Correctly cites timestamps and speaker context |
| Long-context handling | Handles transcripts up to 5 hours without degradation |
What you can ask
AI Chat works like a research assistant who has read every word of your transcript. Example queries that work well:
- "What was the budget number the CFO mentioned?"
- "List all the concerns Sarah raised about the timeline."
- "What action items were assigned to the engineering team?"
- "Did anyone push back on the proposal? What did they say?"
- "Summarize everything said between 14:00 and 22:00."
- "Who spoke the most? What were their main points?"
Follow-up questions work naturally - you don't need to re-state context in every message.
Export and limits
Full AI Chat conversation history can be exported. Monthly message quotas by plan:
| Plan | AI Chat Messages/Month |
|---|---|
| Free | 2 (one-time, not recurring) |
| Starter | 50 |
| Pro (monthly) | 250 |
| Pro (yearly) | 300 |
| Business (monthly) | 750 |
| Business (yearly) | 1,000 |
Transcription Engine: No Change
The transcription engine is unchanged - our core speech recognition model continues to deliver best-in-class accuracy for multi-language, multi-speaker audio. It runs with:
- Automatic language detection across 99+ languages
- Automatic speaker diarization (multi-speaker identification)
- Custom vocabulary support for jargon-heavy domains
The AI model upgrades above apply only to the post-transcription analysis layer - they don't change how audio is converted to text.
What This Means for Active Users
If you use AI Summaries: The first time you generate a summary after this update, you'll see the model selector. Your existing summaries are preserved - new summaries use the upgraded models. You may notice summaries are longer, richer, and more accurately attributed to individual speakers.
If you use AI Chat: Responses are noticeably faster. Answers are more precise on long transcripts. If you previously found AI Chat to be slower or occasionally vague, this update directly addresses that.
If you use neither: Nothing changes for transcription, annotation, or export.
What's Next
Our AI roadmap for Q2–Q3 2026:
- Custom summary templates (Pro/Business) - define your own section names and output format
- Cross-transcript chat - ask questions across multiple recordings in a single query
- Auto-chapter markers - AI-detected topic transitions marked on the transcript timeline
- Summary comparison mode - run both GPT-5.4 and Claude Opus 4.7 and see a side-by-side diff
Questions about the new models? Reach out via the Help Center.