OpenAI ChatGPT snapshot
GPT-5.3 Chat Latest
A deprecated API snapshot of the GPT-5.3 Instant model that powered ChatGPT, useful today mainly as a historical reference for conversational behavior, cost, and migration planning.
- Context window
- 128K
- tokens
- Max output
- 16,384
- tokens
- Input
- $1.75
- per 1M tokens
- Cached input
- $0.175
- per 1M tokens
- Output
- $14.00
- per 1M tokens
01 / Overview
What GPT-5.3 Chat Latest Was
GPT-5.3 Chat Latest was an API alias for the GPT-5.3 Instant snapshot used in ChatGPT, giving developers temporary access to the conversational model behavior associated with ChatGPT rather than a long-lived production model family.
A ChatGPT Snapshot, Not a Stable Production Line
OpenAI released gpt-5.3-chat-latest to the Responses and Chat Completions APIs on March 3, 2026. The alias pointed to the GPT-5.3 Instant snapshot currently used in ChatGPT and was updated as that ChatGPT snapshot changed.
That design made the model useful for developers who wanted to evaluate ChatGPT-style conversational behavior through the API. It also made the alias fundamentally different from a stable production model with a long support horizon.
The model focused on text conversation with optional image input. It supported streaming, function calling, and structured outputs, which made it capable enough for conversational applications that still needed application-controlled responses.
Its lifecycle was short. OpenAI notified developers of deprecation on May 8, 2026 and listed August 10, 2026 as the API shutdown date. The historical migration recommendation was GPT-5.6 Sol, while the current GPT-5.3 Chat documentation recommends GPT-6 Astra for most API usage.
- Alias for the GPT-5.3 Instant snapshot used in ChatGPT.
- Released for API access in March 2026.
- Supported conversational text and image-aware workflows.
- Supported streaming, function calling, and structured outputs.
- Deprecated and scheduled for API shutdown on August 10, 2026.
- Provider
- OpenAI
- Family
- GPT-5.3 Chat
- Positioning
- ChatGPT Instant snapshot
- Status
- Deprecated
- Knowledge cutoff
- Aug 31, 2025
- Input modalities
- Text, Image
- Output modality
- Text
02 / Use cases
Where GPT-5.3 Chat Latest Still Matters
GPT-5.3 Chat Latest is most useful today as a migration and evaluation reference: teams can use its documented behavior and historical test results to understand what must be preserved when moving an older conversational integration to a supported model.
Preserve User Experience While Changing the Model
A migration can be technically successful and still change the product experience. Tone, response length, instruction following, schema reliability, and conversational style can all shift when a new model replaces an older chat snapshot.
That makes representative evaluation important. Existing teams should identify the prompts and conversations that defined acceptable GPT-5.3 behavior, then run the same cases against candidate replacements.
Structured-output workflows deserve their own test set. GPT-5.3 Chat supported structured outputs and function calling, so migration should verify that schemas, tool arguments, validation behavior, and downstream application assumptions still hold.
Vision-assisted chat is another relevant category because the model accepted image input. If an older product relied on screenshots, documents, or visual context inside a conversation, those cases should be included in migration testing.
- Historical conversational regression testing.
- Tone and response-style migration checks.
- Structured-output compatibility evaluation.
- Function-calling regression tests.
- Image-assisted chat migration benchmarks.
- 01
Conversation regression
Compare tone, directness, formatting, and instruction following on the prompts that defined the original user experience.
- 02
Structured output migration
Verify that schemas and machine-readable responses remain compatible after switching models.
- 03
Function-call compatibility
Re-run application tool and function scenarios to identify differences in arguments, sequencing, or completion behavior.
- 04
Vision-assisted chat
Include historical screenshot and image-input cases when the original integration depended on multimodal conversations.
03 / Pricing
GPT-5.3 Chat Latest Pricing
OpenAI's preserved GPT-5.3 Chat model card lists $1.75 per million input tokens, $0.175 per million cached input tokens, and $14.00 per million output tokens.
Treat Pricing as a Migration Baseline
Because the model has been deprecated and its API shutdown date has passed, these prices are most useful as a historical baseline rather than as the economics for a new deployment.
A team migrating an existing integration can compare the old cost profile with a supported replacement. The relevant metric is not only price per million tokens: changes in average response length, retries, context size, and task success rate can alter total cost per conversation.
GPT-5.3 Chat had a ten-to-one difference between standard and cached input pricing. For applications with repeated system instructions or stable conversational prefixes, that could materially reduce the input side of the historical cost profile.
Output was substantially more expensive than input, so verbose conversational behavior could dominate spend even when prompts were relatively small.
- $1.75 per 1M input tokens.
- $0.175 per 1M cached input tokens.
- $14.00 per 1M output tokens.
- Tool-specific models or operations could carry separate fees.
- Use these values as a historical migration baseline rather than a new deployment target.
1M tokens · USD
- Input
- $1.75
- Cached input
- $0.175
- Output
- $14.00
Example: 10K input + 2K output
- Input cost
- $0.0175
- Output cost
- $0.0280
- Estimated total
- $0.0455
04 / Context
A 128K Context Window for Conversational Workloads
GPT-5.3 Chat Latest had a 128,000-token context window and a maximum output of 16,384 tokens, a much smaller working envelope than the million-token windows available on later frontier models.
Context Limits Are Part of Migration Behavior
The smaller context window shaped how applications built around GPT-5.3 Chat managed conversation history.
Long chats often required truncation, summarization, retrieval, or selective message retention before the request approached the 128K limit. Those strategies may still be embedded in older applications even when a replacement model supports far more context.
Migration is therefore an opportunity to test context strategy separately from model quality. A larger context window does not automatically mean that sending the entire history is the best design.
The 16,384-token maximum output is also worth preserving in regression tests. If an application was designed around concise conversational answers, moving to a model with a much larger output allowance can change latency, cost, and product feel unless response length is constrained explicitly.
- 128,000-token context window.
- Maximum output of 16,384 tokens.
- Smaller working context than later GPT-5.4, GPT-5.5, GPT-5.6, and GPT-6 models.
- Historical applications may contain truncation or summarization logic built around this limit.
- Migration tests should isolate model behavior from context-strategy changes.
Context window
128,000
Max output
16,384
Legacy chat applications often managed history aggressively around the 128K context window. Preserve or deliberately re-evaluate that behavior during migration.
05 / Reasoning
No Published Reasoning-Effort Matrix
OpenAI's GPT-5.3 Chat model card does not publish a reasoning_effort level matrix for this chat snapshot, so reasoning controls should not be inferred from newer GPT-5.4, GPT-5.5, GPT-5.6, or GPT-6 models.
Evaluate Observable Behavior Instead of Assuming Hidden Controls
GPT-5.3 Chat was exposed as the Instant model used in ChatGPT rather than as a model family documented around configurable reasoning levels.
For migration work, the most reliable approach is therefore behavioral: identify the old prompts that required careful reasoning, measure the outputs they produced, and compare those results against explicit reasoning settings on a supported replacement where available.
This also prevents a common migration mistake. Newer models may expose controls such as none, low, medium, or higher reasoning levels, but applying one of those settings does not automatically recreate GPT-5.3 Chat behavior.
The target should be task success, latency, cost, tone, and response structure—not a guessed one-to-one mapping of internal inference settings.
Legacy reasoning case
Preserve representative prompts and expected outcomes instead of assuming an undocumented GPT-5.3 reasoning setting.
Replacement model
Test the supported replacement at several documented reasoning levels and select the lowest setting that meets the migration quality bar.
06 / Capabilities
Conversational API Features
GPT-5.3 Chat Latest supported streaming, function calling, structured outputs, text input and output, and image input, making it suitable for conversational applications that needed more than free-form text.
Chat Behavior with Application Control
Streaming allowed applications to render responses progressively, which is especially useful in chat interfaces where perceived latency matters.
Function calling let the model return arguments for application-defined actions, while structured outputs gave developers a stronger contract for machine-readable responses.
Image input enabled multimodal conversations involving screenshots, photos, documents, or other visual material. The model itself produced text output; direct audio and video modalities were not supported.
Fine-tuning and predicted outputs were also listed as unsupported on the model card.
- Text input and text output.
- Image input.
- Streaming supported.
- Function calling supported.
- Structured outputs supported.
- Audio and video not supported.
- Fine-tuning not supported.
- Predicted outputs not supported.
- Supported
Streaming
Return conversational text progressively for responsive chat interfaces.
- Supported
Function calling
Connect model responses to application-defined actions and tools.
- Supported
Structured outputs
Produce machine-readable responses that conform to application schemas.
- Supported
Image input
Include visual context such as screenshots, photos, and documents in a conversation.
- Not listed
Fine-tuning
OpenAI lists fine-tuning as unsupported for GPT-5.3 Chat.
- Not listed
Predicted outputs
OpenAI lists predicted outputs as unsupported for this model.
07 / Evaluation
Strengths and Limitations
GPT-5.3 Chat Latest is most valuable now as a documented snapshot of an earlier ChatGPT Instant experience: useful for migration baselines and historical evaluation, but no longer a suitable target for new API integrations.
Historical strengths
ChatGPT-aligned behavior
The alias pointed directly to the GPT-5.3 Instant snapshot used in ChatGPT, making it useful for testing ChatGPT-style conversational behavior through the API.
Structured application support
Streaming, function calling, and structured outputs supported production-style conversational integrations.
Image-aware chat
Image input allowed applications to combine visual context with text conversations.
Useful migration baseline
Existing prompts, costs, and regression sets built around GPT-5.3 Chat can provide a concrete benchmark for a replacement model.
What to consider
API lifecycle ended
OpenAI's deprecation schedule lists August 10, 2026 as the shutdown date, so the model should not be selected for new API deployments.
Snapshot-style alias
The alias tracked the Instant model used in ChatGPT rather than representing a stable long-lived production family.
Smaller context window
The 128K context and 16K output limits are substantially smaller than those of newer frontier models.
No published reasoning matrix
The current model card does not document configurable reasoning-effort levels for GPT-5.3 Chat.
Plan the migration
Benchmark legacy chat behavior against a current model
Use representative conversational prompts and structured-output cases to document your old GPT-5.3 Chat baseline, then compare quality, token usage, and cost against a supported replacement in EidoStack.
Start FreeGPT-5.3 Chat Latest is no longer an appropriate target for new API deployments. Use historical evaluations to preserve expected behavior while migrating to a supported model.
Common Questions
What is GPT-5.3 Chat Latest?
GPT-5.3 Chat Latest was an OpenAI API alias that pointed to the GPT-5.3 Instant snapshot used in ChatGPT. It was designed to expose ChatGPT-style conversational behavior through the API.
Is GPT-5.3 Chat Latest deprecated?
Yes. OpenAI announced its deprecation in May 2026 and listed August 10, 2026 as the API shutdown date.
What model should replace GPT-5.3 Chat Latest?
OpenAI's May 2026 deprecation notice listed GPT-5.6 Sol as the recommended replacement. The current GPT-5.3 Chat model card recommends GPT-6 Astra for most API usage.
How much did GPT-5.3 Chat Latest cost?
The preserved model card lists $1.75 per 1M input tokens, $0.175 per 1M cached input tokens, and $14.00 per 1M output tokens.
What is the context window of GPT-5.3 Chat Latest?
GPT-5.3 Chat Latest has a documented 128,000-token context window and a maximum output of 16,384 tokens.
What is the knowledge cutoff for GPT-5.3 Chat Latest?
OpenAI lists August 31, 2025 as the knowledge cutoff for GPT-5.3 Chat.
Does GPT-5.3 Chat Latest support image input?
Yes. The model accepts text and image input and produces text output. Direct audio and video modalities are not supported.
Does GPT-5.3 Chat Latest support function calling?
Yes. OpenAI lists function calling as supported.
Does GPT-5.3 Chat Latest support structured outputs?
Yes. Structured outputs are listed as supported on the model card.
Does GPT-5.3 Chat Latest support streaming?
Yes. Streaming is listed as supported.
What reasoning levels does GPT-5.3 Chat Latest support?
OpenAI's GPT-5.3 Chat model card does not publish a reasoning-effort level matrix. Do not assume that reasoning settings from newer GPT model families apply to this deprecated chat snapshot.
Can GPT-5.3 Chat Latest be fine-tuned?
No. OpenAI lists fine-tuning as unsupported for GPT-5.3 Chat.
Why keep a GPT-5.3 Chat page after deprecation?
A historical model page remains useful for migration research, cost comparisons, regression testing, understanding old API integrations, and documenting the behavior of applications originally built around the GPT-5.3 Instant snapshot.
Model information
Last updated
The specifications and lifecycle information on this page are based on OpenAI's GPT-5.3 Chat model card, API changelog, and deprecation documentation. OpenAI announced deprecation in May 2026 and listed August 10, 2026 as the shutdown date. The model card remains available as documentation for the deprecated GPT-5.3 Instant snapshot.