Skip to content
Model infoModelsSnapshot: 15 Oct 2026

OpenAI / GPT-REALTIME

GPT-Realtime

OpenAI speech generation, agents, vision model with developer-documented API access and version information.

Speech & voice
deprecatedClosed weightsSpeech generationAgentsVision
Official documentation (opens in a new tab)Reviewed 12 Oct 2026
11 of 18 specification fields collected
Context window32K32,000 tokens
Maximum output4.1K4,096 tokens
Input / 1M tokens$4.00OpenAI API
Output / 1M tokens$16.00USD · observed 12 Oct 2026

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
32,000 tokensSource for Context / input limit (opens in a new tab)
Maximum output
4,096 tokensSource for Maximum output (opens in a new tab)
Input modalities
Text, Audio, ImageSource for Input modalities (opens in a new tab)
Output modalities
Text, AudioSource for Output modalities (opens in a new tab)
Reasoning
Not collected
Tool calling
SupportedSource for Tool calling (opens in a new tab)
Structured output
Not collected
Streaming
Not collected
Batch processing
Not supportedSource for Batch processing (opens in a new tab)
Fine-tuning
Not supportedSource for Fine-tuning (opens in a new tab)
Knowledge cutoff
2023-09-30Source for Knowledge cutoff (opens in a new tab)
Release date
Not collected
Parameters
Not collected
Architecture
Not collected
Languages
Not collected
Supported API routes
v1/realtimeSource for Supported API routes (opens in a new tab)
Snapshots and aliases
gpt-realtime-2025-08-28Source for Snapshots and aliases (opens in a new tab)

Aliases and pinned IDs listed by the developer. A snapshot date is not automatically the public release date.

Context caching
SupportedSource for Context caching (opens in a new tab)

Documented on at least one recorded endpoint. See Caching for host, lifecycle and billing restrictions.

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

OpenAI API

Provider source (opens in a new tab)
API model ID
gpt-realtime
Access
API account required. Model-specific permissions and rate limits apply.
Regions
Not collected
INPUT / 1M TOKENS$4.00
OUTPUT / 1M TOKENS$16.00

Standard text-token rates. Other modalities, tool usage and processing modes are billed separately. Cached input price: $0.40 / 1M tokens (base tier).

USD · observed 12 Oct 2026 · Pricing evidence (opens in a new tab)
Published billing units · original currency · observed 12 Oct 2026
CategoryUsageRateUnit
Text tokensInput1$4.001M tokens
Text tokensCached input1$0.401M tokens
Text tokensOutput1$16.001M tokens
Audio tokensInput1$32.001M tokens
Audio tokensCached input1$0.401M tokens
Audio tokensOutput1$64.001M tokens
Image tokensInput1$5.001M tokens
Image tokensCached input1$0.501M tokens

Rate conditions

  1. Base published rate. Check the model-card pricing conditions and the request modality.
Pricing source (opens in a new tab)

Data & enterprise terms

API data controls and retention depend on endpoint and account configuration; consult the linked provider policy.

Read the service policy (opens in a new tab)

Caching

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

01 / INITIAL INPUTProcess & retain

The system processes eligible input and retains reusable state.

02 / LATER REQUESTCheck available state

Reuse depends on matching input, session state and the cache policy.

03 / CACHE HITReuse & generate

Reusing state can reduce repeated work. Any charges follow the billing rules below.

OpenAI API

gpt-realtime
Documented
Support evidence (opens in a new tab)
How it works
Automatic prefix reuseSource for How it works (opens in a new tab)
Minimum reusable input
Varies with model and request settings; hidden system tokens do not count toward the minimum.Source for Minimum reusable input (opens in a new tab)
Lifetime & refresh
In-memory entries: typically 5–10 minutes inactive, at most 1 hour. Extended retention is not listed for this model.Source for Lifetime & refresh (opens in a new tab)
How to enable it
Keep the prompt prefix identical. A stable prompt_cache_key can improve routing; it does not guarantee a hit.Source for How to enable it (opens in a new tab)
Usage fields
Inspect cached_tokens in API usage; older models report multiples of 128 after excluding hidden system tokens.Source for Usage fields (opens in a new tab)
Billing rules
No additional cache-write charge. Reused input uses the cached-input rate; uncached input and output retain their normal rates.Source for Billing rules (opens in a new tab)
Limits & data boundaries
Exact rendered-prefix matching. No sharing across organizations or processing regions. Routing and load can prevent hits. Check account retention policy before selecting persistent caching.Source for Limits & data boundaries (opens in a new tab)
Cache reads, writes & storage
ChargeApplies toRate / unit
Cache read1Text tokens · Base tier$0.40 / 1M tokensPricing evidence (opens in a new tab)
Cached input2Audio tokens$0.40 / 1M tokensPricing evidence (opens in a new tab)
Cached input2Image tokens$0.50 / 1M tokensPricing evidence (opens in a new tab)

Rate conditions

  1. Standard text-token rates. Other modalities, tool usage and processing modes are billed separately.
  2. Base published rate. Check the model-card pricing conditions and the request modality.

Observed 12 Oct 2026. Write rates and storage fees are different charges; use the billing rules above.

Caching reviewed 12 Oct 2026

What the measurements say

No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

The record behind the model

Why it is included

Proprietary OpenAI release in the official hosted-model catalog. Explicit open-weight releases and unresolved routing aliases are excluded.

Classification evidence (opens in a new tab)

Lifecycle dates

Deprecation announced: gpt-realtime20 Jul 2026Source (opens in a new tab)
API shutdown: gpt-realtime19 Jan 2027Source (opens in a new tab)

Sources reviewed