Sign in
GPT-Live-1 logo

GPT-Live-1

OpenAI · Released Jul 2026

View pricingCompare models
Conditional

GPT-Live-1 is OpenAI's full-duplex voice model: it listens and speaks at the same time and hands reasoning and tool use to a backend model. Since 10 September 2026 it is generally available in the API on the Live endpoint at $0.05 per minute, with backend model and tool usage billed separately. Conditional because its benchmarks are OpenAI's own, language quality is uneven, and real cost per call depends on the backend model you pair it with.

Is it right for you?

Good for

  • Natural, interruptible voice conversations: full-duplex architecture eliminates turn-based rigidity
  • Complex Q&A during voice: delegates reasoning and search to a backend model without dropping conversation flow
  • Production voice agents via the API: generally available since 10 September 2026 on the Live endpoint (v1/live/sessions) at $0.05 per minute, billed per second, with streaming and function calling
  • Hands-free long-form interaction: designed for 30+ minute conversations, commute use, language practice
  • Real-time voice translation: English to Hindi, Spanish, and other languages with low latency

Not good for

  • Budgeting on the per-minute rate alone: OpenAI bills backend model and tool usage separately from the $0.05/min voice session, so real cost per call depends on the model doing the reasoning
  • Stacks built on Chat Completions, Responses or the Realtime API: gpt-live-1 runs only on the Live endpoint, and structured outputs and fine-tuning are not supported, so it is not a drop-in model swap
  • Languages beyond the most popular: non-native accents and fluency gaps reported (e.g., Hindi)

How it performs by task

Voice conversation

Very Good

Strongly preferred over Advanced Voice Mode in human evals; full-duplex is a real jump

Real-time translation

Good

Functional with low latency, but accent quality varies significantly by language

Scientific reasoning (voice)

Very Good

Substantially outperforms Advanced Voice Mode on GPQA via GPT-5.5 delegation

Web search (voice)

Very Good

Strong gains over Advanced Voice Mode on BrowseComp agentic search benchmark

Code generation

Poor

Not designed for coding — voice-native model; delegates to GPT-5.5 for complex work

Pricing

Input

$0.05 / minute (voice session)

Output

Included in per-minute session rate

Context

Live API; backend model + tools billed separately

View full pricing

Benchmarks

BenchmarkScoreSource
GPQA (scientific reasoning, voice)Substantially outperforms Advanced Voice Mode Source
BrowseComp (agentic web search, voice)Strong gains over Advanced Voice Mode Source
tau^3-Voice Telecom (voice agent tasks)Outperforms Advanced Voice Mode Source

Verdict history

Sep 26, 2026
conditional -> conditional (summary corrected: API generally available): rating_before=conditional, rating_after=conditional. The standing summary said GPT-Live-1 was 'a product-integrated model with no API'. OpenAI's API changelog (Sep 10, 2026) states 'GPT-Live 1 became generally available in the API', and the model page, fetched 26 September 2026, states 'Voice sessions cost $0.05 per minute, billed per second' with backend model and tool usage charged separately, Live endpoint only. The entity's own pricing fields and strengths already carried this since 2026-09-11; staged payload B never landed because the announcing post published without it. Rating held: benchmarks remain OpenAI-reported and language support is uneven.
Jul 15, 2026
copy edit to brevity standard: detail preserved in existing structured children

Sources

Verification log

  • Pricing· No changes

    Automated agent

  • Pricing· No changes

    Automated agent

  • Pricing· No changes

    Automated agent

  • Pricing· No changes

    Automated agent

  • Profile· Updated

    Automated agent

    Pricing unchanged: model page 'Voice sessions cost $0.05 per minute, billed per second', backend model and tools billed separately, Live endpoint only; pricing page row gpt-live-1 $0.05/min. Scalars match. verdict_summary said 'product-integrated model with no API' against API GA of Sep 10 (changelog). Staged payload B never landed: announcing post gpt-live-1-api-launch-voice-layer-pricing is already published without it. Corrected via standalone verdict_set 53b52a60-d1e4-4b77-9d33-2af3f5dadbb4 linked to that post. rating_before=conditional, rating_after=conditional. No model_update, no new draft (fact already announced).

  • Pricing· No changes

    Automated agent

    gpt-live-1 '$0.05 per minute'. Matches stored scalars. No update. verdict_summary still says 'no API', which contradicts its own strengths (Live API GA 10 Sep). The fix is payload B on draft gpt-live-1-api-launch-voice-layer-pricing and has not landed (verdict_history latest 2026-07-15). Not re-staged and no verdict_set, so the one change does not get two paths.

  • Pricing· No changes

    Automated agent

    Pricing row gpt-live-1 'Price per minute: $0.05'. Matches stored scalars. No update. Payload B (same-rating, API availability) on draft gpt-live-1-api-launch-voice-layer-pricing still NOT landed: summary still says 'no API', verdict_history latest 2026-07-15. Not re-staged.

  • Pricing· No changes

    Automated agent

    Model page verbatim: 'Voice sessions cost $0.05 per minute, billed per second. Backend model and tool usage is billed separately.' Live endpoint only (v1/live/sessions); streaming + function_calling; structured_outputs, fine_tuning, predicted_outputs unsupported; Tier 1-5 concurrent sessions 25/50/200/300/500; Free tier unsupported. Pricing page row 'Price per minute $0.05'. Matches stored scalars and children. No update. Staged same-rating payload B (summary still says 'no API') has NOT landed: verdict_history latest 2026-07-15. Not re-staged.

  • Pricing· Updated

    Automated agent

    MATERIAL: the entity said 'no API' and priced the model as 'ChatGPT subscription'. OpenAI API changelog, Sep 10 2026, verbatim: 'GPT-Live 1 is now generally available in the API. Build full-duplex voice conversations that can continue while a backend model or agent handles reasoning and tools.' Model page, verbatim: 'Voice sessions cost $0.05 per minute, billed per second. Backend model and tool usage is billed separately.' Endpoint: Live only (v1/live/sessions); Chat Completions, Responses, Realtime, Assistants, Batch and the rest 'Not supported'. Features: streaming, function_calling; not structured_outputs or fine_tuning. Rate limits Tier 1-5 = 25/50/200/300/500 concurrent sessions; free tier unsupported. Pricing page row 'gpt-live-1 Price per minute $0.05' corroborates. REFRESHED: pricing_input 'ChatGPT subscription' -> '$0.05 / minute (voice session)'; pricing_output -> 'Included in per-minute session rate'; pricing_context 'ChatGPT Voice only' -> 'Live API; backend model + tools billed separately'; pricing_url openai.com/chatgpt/pricing/ -> developers.openai.com/api/docs/pricing; replaced the two now-false avoid_for rows (no API / no standalone pricing, no SDK) with an API best_for row plus separate-billing and Live-endpoint-only avoid_for rows; replaced the 'API developers' not_for_whom audience with a for_whom developer audience and a free-tier API not_for_whom. The 'GPT-5.5' backend claim was genericized in strengths because the API page does not name the backend model (task_strengths/benchmarks untouched, they cite the July launch post). VERDICT: rating stays conditional (OpenAI-only benchmarks, nothing moves it), but the live summary says 'no API'. Same-rating entity_verdict STAGED for the existing curator draft 'gpt-live-1-api-launch-voice-layer-pricing'; no new draft, no verdict_set. Tool defect re-observed: sources rows for the changelog and pricing URLs resolved idempotently to old labels/access dates (2026-09-04, 2026-06-24).

  • Pricing· No changes

    Automated agent

  • Pricing· No changes

    Automated agent

  • Pricing· No changes

    Automated agent

  • Pricing· No changes

    Automated agent

  • Profile· No changes

    Imported at launch

  • Pricing· No changes

    Imported at launch

How we evaluate