AIREITERA model can be officially available without having a public rate card. That is the current Gemini 3.8 Live situation: Google announced Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking on September 15, 2026, but its documentation still does not give a separate 3.8 Live dollar price. Do not copy Gemini 3.8 Flash’s $0.75/$3.75 token rates into a voice-agent budget.
The short answer: Gemini 3.8 Live is real, but its API price is not published yet
Google’s launch post says both models began rolling out to developers through the Gemini API and Google AI Studio on September 15, 2026. That establishes availability, not a finalized public price. The announcement calls Gemini 3.8 Live cost-efficient and Extended Thinking competitively priced, but it does not state a per-minute or audio-token rate.
| Question | Answer |
|---|---|
| Is Gemini 3.8 Live officially launched? | Yes, Google announced it on September 15, 2026. |
| Can developers access it? | Google says rollout began through Gemini API and AI Studio. |
| Is there an official 3.8 Live rate card? | Not on the Gemini API pricing page reviewed for this guide. |
| Is Gemini 3.8 Flash pricing the same thing? | No. gemini-3.8-flash and the Live models are separate model identifiers. |
| Can you calculate an exact 3.8 Live call cost today? | No, not without an account-specific price or later official rate entry. |
What Google launched on September 15, 2026
Google introduced two live-dialogue audio models with different operating goals. Gemini 3.8 Live targets fluid, scalable conversations; Gemini 3.8 Live Extended Thinking targets harder multi-step work while maintaining the conversation.
The launch announcement describes near-real-time visual input, automatic switching across 97 languages, and background tool or API calls that can continue while the model acknowledges the user. Extended Thinking is described as reasoning and speaking simultaneously, with progress cues during longer tasks. These features matter to pricing because background work, tool calls, and deeper reasoning can change token usage.
What access is available to API buyers?
Google’s rollout is not one uniform switch. The official announcement lists the practical developer paths as follows:
| Surface | Gemini 3.8 Live | Gemini 3.8 Live Extended Thinking |
|---|---|---|
| Gemini API | Rollout began September 15, 2026 | Rollout began September 15, 2026 |
| Google AI Studio | Rollout began September 15, 2026 | Rollout began September 15, 2026 |
| Gemini Enterprise | Private preview | Private preview |
Treat that table as Google’s rollout statement, not a guarantee that every account sees the same model ID immediately. The API model list and your project permissions are the operational checks.
How Gemini Live API billing works before a 3.8 rate card exists
Google’s Live API best-practices documentation describes token-based billing rather than a flat session fee. Audio is converted into audio-token usage; text messages and generated transcriptions can create additional text-token usage. A long-running session can also become more expensive as active conversation context accumulates.
The important budgeting rules are:
- Do not treat minutes connected as the official billing unit. A per-minute number is an estimate derived from tokenization and speaking time.
- Separate input audio from output audio. The model can listen for ten minutes and speak for four; those are different usage streams.
- Account for context growth. Google’s guidance warns that accumulated native-audio context can be included in billing on later turns, subject to context handling.
- Add transcription separately when enabled. Transcripts can introduce text-token charges in addition to audio usage.
- Track tool calls and reasoning tokens. Background work can use more output or context tokens than a simple turn-taking bot.
This is why “half a cent per minute” is not a substitute for an official 3.8 price. It may describe one audio direction under one tokenization assumption.
What you can budget today
Until Google publishes Gemini 3.8 Live rates, use currently priced Live models as a planning baseline—not as a proxy for the new model’s price. The official Gemini pricing documentation lists these reference rates:
| Model | Audio input | Audio output | Text input | Text output |
|---|---|---|---|---|
| Gemini 3.1 Flash Live Preview | $3.00 per 1M audio tokens (about $0.005/min) | $12.00 per 1M audio tokens (about $0.018/min) | $0.75 per 1M | $4.50 per 1M |
| Gemini 2.5 Flash Native Audio | $3.00 per 1M audio tokens | $12.00 per 1M audio tokens | $0.50 per 1M | $2.00 per 1M |
| Gemini 3.5 Live Translate | $3.50 per 1M audio input tokens | $21.00 per 1M audio output tokens | Not listed | Not listed |
For a planning example, assume ten minutes of incoming audio and four minutes of model speech using the published Gemini 3.1 Flash Live estimates:
- Input: 10 × $0.005 = $0.05
- Output: 4 × $0.018 = $0.072
- Audio subtotal: $0.122
That is a reference calculation, not a Gemini 3.8 Live quote. It excludes accumulated context, transcription, tools, and any new 3.8 pricing. The minute conversions are estimates, so use token usage from your own billing response for a final budget.
Standard Gemini 3.8 Flash is not the Live API model
Google’s standard gemini-3.8-flash is a text-output model with multimodal input. The official pricing page lists introductory rates of $0.75 per 1 million input tokens and $3.75 per 1 million output tokens through December 31, 2026, with scheduled rates of $1.50/$7.50 from January 1, 2027.
Those prices apply to the standard Flash model, not automatically to gemini-3.8-live or gemini-3.8-live-extended-thinking. The model documentation treats standard Flash separately from Live capability. The names share a version number, but Live needs streaming audio output, turn detection, interruption handling, and real-time session behavior.
A safe implementation should log the exact model ID returned by the provider. “Gemini 3.8” in a dashboard is not precise enough for cost attribution.
Should you build on Gemini 3.8 Live now?
Build a prototype now if your main question is interaction design: can users tolerate background work, verbal progress updates, visual grounding, and interruptions in your workflow? Google has announced those behaviors, and the API/AI Studio rollout gives developers a way to test the integration.
Delay a production cost commitment if your business case depends on a fixed per-minute margin. The missing 3.8 rate card, unknown context-rebilling impact, and absence of independently measured latency and failure data leave too many variables for a reliable unit-economic forecast.
Create a cost harness before switching traffic:
- Record exact model IDs and API responses.
- Capture input audio minutes, output audio minutes, text tokens, thoughts tokens when exposed, transcription, and tool calls.
- Run short, medium, and long sessions with the same task set.
- Compare cost per successfully completed task, not cost per connected minute.
- Keep a fallback on a currently priced Live model until Google publishes the 3.8 rate card.
Gemini 3.8 Live API FAQ
What is the official Gemini 3.8 Live API price?
Google’s launch announcement does not state a dollar price, and the reviewed pricing documentation does not provide a distinct 3.8 Live rate card.
Is Gemini 3.8 Live free in Google AI Studio?
Google says rollout began in AI Studio, but access, quotas, and free-tier terms are separate from a guaranteed production API price. Check the account-level model availability and billing page.
Is Gemini 3.8 Flash the same as Gemini 3.8 Live?
No. gemini-3.8-flash is a separate standard model. Its $0.75/$3.75 introductory token prices should not be presented as the price of a Live audio session.
Does Gemini Live API charge per minute?
Not as its fundamental billing rule. Google documents token-based billing; per-minute figures are estimates based on audio-token rates and assumed speaking duration.
Can a long Live conversation cost more over time?
Yes. Active conversation context, audio tokens, transcription, and tool activity can affect later-turn usage. Measure a representative session rather than multiplying one short-call estimate.
What is the difference between Gemini 3.8 Live and Extended Thinking?
Google positions Gemini 3.8 Live for fluid, scalable dialogue and visual grounding. Extended Thinking is for deeper multi-step reasoning while continuing to speak and report progress.
Where is Gemini 3.8 Live available?
Google announced rollout through Gemini API and AI Studio, with enterprise access initially described as private preview. Availability can differ by account and region.
The defensible budget today is therefore a range built from current Live token rates and measured session behavior—not a guessed Gemini 3.8 Live per-minute price. Revisit the calculation when Google adds the model’s official rate entry.