Service eligibility and regional restrictions
PuppyIP serves only compliant overseas businesses and their authorized personnel. Proxy services are not available in mainland China. The service may only be used for lawful business activities outside mainland China. Use of this service within mainland China is prohibited.
Hosting a proxy IP or server overseas does not change these restrictions. The service must not be provided to end users in mainland China through relaying, forwarding, sharing or resale. Before use, read the Terms of Service.
Key Takeaways
- The enterprise Live Avatar GA announcement provides US and EU endpoints. Custom avatar creation still needs enterprise allowlisting and verification; enterprise Extended Thinking remains private preview.
- Developer API gemini-3.8-live and gemini-3.8-live-extended-thinking, personal Search/Gemini Live, Workspace subscriptions and enterprise Live Avatar are separate entry points. Enterprise GA does not authorize every account, feature or region.
- On September 28, Developer API Standard audio input was $3 and audio output $12 per million tokens. These are API prices, not an enterprise Avatar quote or fixed conversation total.
- The pricing page says free-tier data is used to improve products and paid-tier data is not. Check applicable processing terms separately. PRODUCT_FIT=NONE.
Choose the task before choosing Live or Extended Thinking
The initial Google announcement is dated September 15, 2026 and describes rollout through separate channels. The Developer API model list gives Stable identifiers gemini-3.8-live and gemini-3.8-live-extended-thinking. Live targets low-latency voice interaction; Extended Thinking adds background reasoning. The September 25 enterprise Avatar GA is a separate scope update, not blanket access to both API models or a same-name replacement for 3.8 Flash text calls.
For a hypothetical support assistant answering simple questions, assess Live first. For one comparing multiple constraints during conversation, consider Extended Thinking as well. This is a selection example, not a performed test. Use the same task set to compare correct completion, interruption handling, tool-result consistency and actual bills instead of relying on launch rankings.
Four product entry points have different qualifications
The September 15 announcement describes Gemini API and AI Studio rollout and then-private-preview Gemini Enterprise. The September 25 Cloud announcement upgrades Gemini 3.8 Live with Live Avatar in Enterprise to GA, with US/EU endpoints and provisioned throughput, while explicitly retaining private preview for enterprise Extended Thinking. Customer Experience was still coming soon in the initial announcement. Personal Search/Gemini Live and Workspace subscriptions have separate eligibility. Avatar GA cannot establish access for all enterprise features, subscribers or regions.
Check the product entry point, account eligibility, rollout and organization policy before selecting a model. Curated avatars differ from custom avatar creation requiring enterprise allowlisting and identity verification. Avatar GA does not automatically include custom avatars or Extended Thinking. Stable API identifiers do not prove feature availability in personal apps. Check the relevant product notice rather than changing IPs, borrowing accounts or bypassing administrators.
Separate API input, output, modalities and search costs
On September 28, the Developer API Standard table for both models listed per-million-token prices of $0.75 for text input, $3 for audio input, $1 for image/video input, $4.50 for text output including thinking, and $12 for audio output. It also listed corresponding rates of $0.005/minute audio input, $0.018/minute audio output and $0.002/minute image/video input. These are Developer API meters, not an enterprise Avatar quote, subscription fee or fixed mixed-conversation total.
Google Search grounding lists 5,000 free requests monthly, shared with Gemini 3.x, then $14 per thousand. Budget incoming audio, model output, other inputs and search separately and check actual metering. Free-tier support is not unlimited quota. No audio generation, paid API call or account bill was tested here; use the actual project quotas.
Data conditions matter as much as trial prices
Under Used to improve our products, the pricing page lists Yes for free tier and No for paid tier. Check applicable terms and organization policy before submitting customer calls, business video or internal materials. That paid-tier entry promises neither zero retention nor unrestricted suitability for sensitive data.
Live API differs from ordinary text requests: its docs describe low-latency audiovisual interaction over WebSocket sessions. This article explains model selection rather than SDK integration. No key, session or tool setup was executed, so no measured latency, failure rate or rollback can be reported. Follow the official Live API guide for implementation.
Keep eligibility, task and cost acceptance separate
Record three outcomes separately. Seeing the model establishes only access; task acceptance checks answers, tools and complete results; cost acceptance uses representative sessions and actual metering rather than demo minutes. For orders, reservations or external writes, reconcile results after interruption or disconnection instead of treating a voice acknowledgement as success. This is advice, not testing performed for a team.
For 429, 503 or configuration errors, consult the Gemini API quota/error-layer guide and API configuration checklist. These general explanations replace neither Live API parameters, account permissions nor official quotas, and grant no unavailable features.
What remains unverified
No account-by-region access, custom-avatar approval, business latency, language quality or running cost was verified here. The Cloud article mentions 97 languages, but target-language and entry-point acceptance still matter. Developer API pricing is not an enterprise contract price. September 25 is the GA announcement date; each account activation and final rollout completion remain unknown.
Revisit this topic when Google changes model status, prices, subscriptions, Avatar GA or Extended Thinking preview scope. This is an EXPLAINER; implementation steps and rollback are not applicable. An implementation guide would need SDK, permission and session validation plus stop conditions. PRODUCT_FIT=NONE: network exits cannot expand model access, free quotas, avatar allowlisting or enterprise preview eligibility.
Sources
Frequently Asked Questions
Is Extended Thinking best for every voice task?
No. It emphasizes background reasoning, while Live emphasizes low latency. Compare completion, interruption response, result consistency and fees on the same tasks. This article reports no performance test.
Does enterprise Avatar GA include custom avatars and Extended Thinking?
Not automatically. The September 25 GA announcement retains allowlisting and verification for custom creation, and private preview for enterprise Extended Thinking.
Does API Stable mean enterprise and personal apps have access?
No. Developer API status, personal rollout, subscription eligibility and enterprise Avatar GA are separate conditions. Check the actual product and account.
Does per-minute audio pricing cover a conversation or enterprise Avatar?
No. Input audio, audio/text output, other inputs and search may be metered separately. A single minute-based rate is not a conversation total or enterprise contract quote.
Can I upload customer calls directly on free tier?
Check organization policy and applicable terms first. Free-tier data is used for product improvement; paid-tier data is not, but that does not promise zero retention or authorize all sensitive data.