Last updated: June 10, 2026
Trace Models & Costs.
Trace currently offers Gemini 3.1 Flash-Lite and Gemini 3.5 Flash. This page explains recommended use cases, audio token estimates, transcript output, and the full cost model for meeting-note analysis.
The two models Trace uses
Gemini 3.1 Flash-Lite
- Model code
gemini-3.1-flash-lite- Input / output limits
- 1,048,576 input tokens / 65,536 output tokens
- Best for
- General meetings, longer audio, and low-cost first-pass transcripts.
Gemini 3.5 Flash
- Model code
gemini-3.5-flash- Input / output limits
- 1,048,576 input tokens / 65,536 output tokens
- Best for
- Long meetings, strategic discussions, domain terms, and more complete meeting notes.
How costs are calculated
Gemini audio understanding is commonly estimated by duration: one second of audio is roughly 32 input tokens. One minute is about 1,920 tokens, and one hour is about 115,200 tokens. Trace then uses the transcript, meeting background, and templates to produce meeting notes.
Longer audio may be processed in stages. When the same audio segment is referenced again during the workflow, Google billing may treat it as cached audio input. This page uses that scenario for common estimates and also shows a conservative upper bound when cached input does not apply.
Transcription cost = normal audio input + possible cached audio input + transcript text input + transcript JSON output.
Analysis cost = (full transcript + meeting background / template / analysis instructions) x analysis model input price + meeting-note output tokens x analysis model output price.
Total cost = transcription cost + analysis cost
Cost estimate assumptions
Flash-Lite audio input USD 0.50 / 1M, cached audio USD 0.05 / 1M, output USD 1.50 / 1M. 3.5 Flash input USD 1.50 / 1M, output USD 9.00 / 1M.
Estimated at 32 audio tokens per second. Longer audio may be processed in segments. Reusing the same audio in the workflow may create cached audio input.
Estimated from a 1-hour meeting with about 36 to 42 minutes of speech, including timestamps, speaker labels, and structured output.
1-hour audio input
Transcript JSON output
input / output tokens
The meeting-note analysis estimate uses 20,000 input tokens and 6,000 output tokens. Interview-style meetings, fast multi-speaker discussions, or dense transcripts may be higher. Presentations with more silence or shorter output may be lower. The tables below are practical estimates for reading, while your actual bill is determined by Google. NTD conversion uses about NTD 30 to 33 per USD.
1-hour meeting workflow breakdown
The example below uses "3.1 Flash-Lite transcription + 3.5 Flash analysis". Actual results vary by speaking speed, language, number of speakers, meeting background, template length, output detail, cached input behavior, and Google's latest pricing.
| Step | Estimated tokens | Calculation | Estimated cost |
|---|---|---|---|
| Audio input | 115,200 normal audio + 115,200 cached audio | Reused audio segment estimated with cached audio input. | About USD 0.063 (NTD 1.9-2.1) |
| Transcript text input | About 4,000 input tokens | Meeting background, time context, and necessary segment context. | About USD 0.001 (under NTD 0.1) |
| Transcript JSON output | About 18,000 output tokens | Full transcript, timestamps, speaker labels, and JSON structure. | About USD 0.027 (NTD 0.8-0.9) |
| Meeting-note analysis | About 20,000 input / 6,000 output tokens | Turns transcript, background, template, and analysis instructions into decisions, risks, action items, and formal notes. | About USD 0.084 (NTD 2.5-2.8) |
| Basic total | 1-hour meeting | 3.1 Flash-Lite transcription + 3.5 Flash analysis, including cached audio input. | About USD 0.175 (NTD 5.3-5.8) |
Cost examples by audio length
| Audio length | Transcription flow | Transcript / analysis tokens | With cached audio input | No cached input |
|---|---|---|---|---|
| 30 minutes | Short-audio flow | 9,000 transcript output; 11,000 / 3,000 analysis tokens | About USD 0.086 (NTD 2.6-2.8) | About USD 0.086 (NTD 2.6-2.8) |
| 60 minutes | Standard long-audio flow | 18,000 transcript output; 20,000 / 6,000 analysis tokens | About USD 0.175 (NTD 5.3-5.8) | About USD 0.227 (NTD 6.8-7.5) |
| 120 minutes | Longer-audio flow | 36,000 transcript output; 38,000 / 12,000 analysis tokens | About USD 0.348 (NTD 10.4-11.5) | About USD 0.451 (NTD 13.5-14.9) |
1-hour model-combination comparison
| Combination | Usage | Estimated total | Tradeoff |
|---|---|---|---|
| All Flash-Lite | Transcription and analysis both use gemini-3.1-flash-lite |
About USD 0.105 (NTD 3.2-3.5) | Lowest cost, good for general short meetings or first drafts. Analysis is more conservative. |
| Sample combination | 3.1 Flash-Lite transcription, 3.5 Flash analysis | About USD 0.175 (NTD 5.3-5.8) | Keeps transcription low-cost while using a stronger model for meeting-note organization. |
| All 3.5 Flash | Transcription and analysis both use gemini-3.5-flash |
About USD 0.442 (NTD 13.3-14.6) | Highest cost, best for domain-heavy or complex meetings when budget allows. |
Other costs: meeting background and review
When importing audio, you can add meeting background such as customer names, products, topics, or common acronyms. This context helps the model recognize proper nouns, people, products, and recurring terms, making the transcript closer to the actual meeting.
This context adds a small number of text input tokens, but the cost is usually low. At 2,000 input tokens, Flash-Lite text input is about USD 0.0005, under NTD 0.02. If it goes into 3.5 Flash analysis, it is about USD 0.003, around NTD 0.09 to 0.10.
Fact-checking cost: If you run an additional fact-checking pass, a 1-hour meeting can add roughly USD 0.051 (NTD 1.5-1.7).
Cached audio input on this page is only used to estimate cost range. Trace does not create a backend cache for users. If Google calculates repeated audio references or cache hits differently, use your Google bill and the official pricing page as the source of truth.