OpenAI model snapshot
GPT-4o (2024-11-20)
A dated GPT-4o snapshot for teams that need reproducible text-and-image behavior, stable structured outputs, and a fixed multimodal baseline for production or migration testing.
- Context window
- 128K
- tokens
- Max output
- 16.4K
- tokens
- Input
- $2.50
- per 1M tokens
- Cached input
- $1.25
- per 1M tokens
- Output
- $10.00
- per 1M tokens
01 / Overview
What GPT-4o (2024-11-20) Is
GPT-4o (2024-11-20) is a dated GPT-4o snapshot. Its API identifier, gpt-4o-2024-11-20, lets an application target a specific GPT-4o version instead of relying only on the moving gpt-4o alias.
A Fixed Multimodal Model Version
The value of this model ID is repeatability. If a production workflow depends on image interpretation, tool selection, structured output, or prompt-specific behavior, changing the underlying model revision can make regressions harder to diagnose.
Pinning the November 2024 snapshot removes that variable. The same screenshot, document image, prompt, schema, and tool definition can be evaluated again later against the same GPT-4o version.
The snapshot keeps the GPT-4o family profile documented by OpenAI: text and image input, text output, a 128,000-token context window, up to 16,384 output tokens, and support for streaming, function calling, structured outputs, fine-tuning, and predicted outputs.
- Fixed API model ID:
gpt-4o-2024-11-20. - Designed for reproducible GPT-4o evaluations and deployments.
- Supports both text and image input.
- Useful as a stable baseline before changing prompts or model families.
- Provider
- OpenAI
- Family
- GPT-4o
- Snapshot
- 2024-11-20
- Model ID
- gpt-4o-2024-11-20
- Knowledge cutoff
- Oct 1, 2023
- Input modalities
- Text, Image
- Output modality
- Text
02 / Why pin it
Why Pin GPT-4o to the 2024-11-20 Snapshot?
A snapshot is useful when the model itself must remain constant while the rest of the application changes.
Make Model Version Part of the Test Environment
A production AI system contains many moving pieces: prompts, retrieval logic, tool definitions, schemas, source data, client code, and the model. If several of those variables change at the same time, it becomes difficult to explain why an output changed.
Using gpt-4o-2024-11-20 gives the model layer a fixed identity. You can modify a prompt, introduce a new function, alter a JSON schema, or change image preprocessing while keeping the GPT-4o revision constant.
This is particularly useful for teams maintaining eval suites or acceptance tests. A regression can be compared against a stable model baseline rather than against an alias whose underlying behavior may change over time.
- 01
Lock the model ID
Use gpt-4o-2024-11-20 explicitly in the test or production configuration.
- 02
Capture a baseline
Save quality, token usage, cost, schema validity, vision accuracy, and tool-call behavior for representative requests.
- 03
Change one system variable
Iterate on prompts, image processing, tools, retrieval, or response schemas without also changing the model snapshot.
- 04
Upgrade deliberately
Compare a newer model or snapshot against the baseline before moving production traffic.
03 / Vision testing
A Stable GPT-4o Baseline for Vision Workloads
The November 2024 snapshot is especially useful when visual behavior needs to be repeatable across screenshot, document-image, chart, and multimodal test cases.
Visual Regressions Can Be Harder to Diagnose Than Text Regressions
A text-only test may fail because wording changes. A vision workflow has additional variables: image resolution, crop, compression, layout, text density, visual hierarchy, and where relevant evidence appears in the image.
If the underlying model also changes, debugging becomes even harder. Pinning GPT-4o lets you test whether changes in image preprocessing or prompts actually improve the result.
Typical evaluation sets can include UI screenshots, scanned forms, invoices, dashboards, diagrams, charts, presentation slides, or product images. The goal is not simply to prove that GPT-4o accepts images, but to measure whether the exact visual task remains reliable.
The standard GPT-4o model entry supports text and image input with text output. Native audio and video are not part of this model entry.
- 01
Screenshot QA
Run the same UI screenshots through stable prompts to detect changes in visual interpretation.
- 02
Document extraction
Measure whether fields, labels, and visual structure are extracted consistently from scanned or photographed pages.
- 03
Chart interpretation
Keep the model fixed while testing prompt changes for charts, plots, dashboards, and diagrams.
- 04
Image preprocessing
Compare crops, compression, detail settings, or layout changes against the same model version.
04 / Pricing
GPT-4o (2024-11-20) Pricing
The GPT-4o November 2024 snapshot uses the GPT-4o pricing profile: $2.50 per million standard input tokens, $1.25 per million cached input tokens, and $10.00 per million output tokens.
A Fixed Model Version Does Not Mean a Fixed Request Cost
Snapshot pinning controls model identity. It does not control how many tokens an application sends or receives.
Vision-heavy requests can consume input tokens differently from text-only prompts, while long structured responses can increase output spend quickly. Cached input can reduce the price of repeated prompt prefixes, stable instructions, or other reusable context.
For regression testing, record cost together with quality. If a prompt revision improves accuracy but doubles output length, that change is part of the production tradeoff.
- $2.50 per 1M standard input tokens.
- $1.25 per 1M cached input tokens.
- $10.00 per 1M output tokens.
- Image-based requests should be measured using real production image inputs.
1M tokens · USD
- Input
- $2.50
- Cached input
- $1.25
- Output
- $10.00
Example: 12K input + 1.5K output
- Input cost
- $0.0300
- Output cost
- $0.0150
- Estimated total
- $0.0450
05 / Context
GPT-4o (2024-11-20) Has a 128K Context Window
The snapshot supports 128,000 tokens of total context and up to 16,384 output tokens, giving it substantial capacity for conventional chat, vision, document, and tool workflows.
Keep Context Stable When You Compare Prompt Changes
A reproducible evaluation depends on more than the model ID. If the system prompt, conversation history, retrieved passages, or number of images changes between runs, the comparison may no longer isolate the variable you intended to test.
With a fixed snapshot, teams can build controlled context experiments. For example, compare two retrieval strategies against the same GPT-4o version, or measure whether including additional screenshot context actually improves support answers.
The 128K window is much smaller than newer million-token models. That makes context selection a more visible design constraint, which can be useful when testing whether a migration to a larger-context model would materially improve the workflow.
- Context window: 128,000 tokens.
- Maximum output: 16,384 tokens.
- Suitable for stable chat, document, image, and tool-call evaluations.
- Useful as a baseline before testing a model with a much larger context window.
Context window
128,000
Max output
16,384
The context budget can include instructions, conversation history, text, image representations, retrieved material, tool results, and generated output.
06 / Capabilities
API Capabilities of the November 2024 GPT-4o Snapshot
The snapshot combines GPT-4o's multimodal input with production API features such as streaming, function calling, structured outputs, fine-tuning, and predicted outputs.
Useful for Stable Integration Tests
Function calling can be evaluated against a fixed set of tools. Structured outputs can be tested against stable JSON schemas. Fine-tuned workflows can use a known GPT-4o family baseline, while predicted outputs can support editing-style cases where most of the expected result is already known.
This makes the snapshot valuable for integration testing, not only response-quality testing. A team can verify whether a prompt or schema revision changes tool selection, field population, response validity, or output verbosity without changing the base model version.
- Supported
Text
Accept text input and produce text output.
- Supported
Image input
Analyze screenshots, photos, document images, diagrams, and other supported visual inputs.
- Supported
Streaming
Receive generated output progressively.
- Supported
Function calling
Generate calls to application-defined functions.
- Supported
Structured outputs
Constrain responses to a defined machine-readable structure.
- Supported
Fine-tuning
Use supported GPT-4o fine-tuning workflows for specialized behavior.
- Supported
Predicted outputs
Optimize supported generation when much of the expected output is already known.
07 / Migration baseline
Use GPT-4o (2024-11-20) as a Migration Baseline
A dated GPT-4o snapshot is useful when a team wants to prove that a newer model is actually better for its own application rather than assuming that newer always means safer to deploy.
Compare Production Behavior, Not Marketing Labels
A migration test should reproduce the workloads that matter: screenshots, image extraction, structured JSON, tool calls, long conversations, edge cases, and prompts that have historically caused failures.
Run those requests against gpt-4o-2024-11-20, then run the same evaluation against the candidate replacement. Compare correctness, visual interpretation, schema validity, latency, output length, token cost, and failure rate.
This gives GPT-4o a useful role even when it is not the model you plan to use long term. It becomes a fixed reference point that makes migration gains and regressions measurable.
- 01
Freeze the GPT-4o baseline
Keep the November 2024 snapshot, prompts, tools, schemas, and evaluation dataset fixed.
- 02
Run the candidate model
Use equivalent application conditions so the model itself is the main changed variable.
- 03
Compare regressions and gains
Inspect vision accuracy, JSON validity, tool behavior, latency, token usage, and cost.
- 04
Migrate based on evidence
Adopt the replacement when measured improvements outweigh compatibility and operational risks.
08 / Evaluation
GPT-4o (2024-11-20) Strengths and Limitations
The snapshot's main strength is controlled repeatability for a mature multimodal model. Its main limitation is that pinning deliberately keeps you on an older model revision until you choose to migrate.
Strengths
Reproducible multimodal behavior
A dated model ID creates a stable reference for screenshot, document-image, and mixed text-image evaluations.
Stable structured workflows
Tool calls and structured outputs can be tested without simultaneously changing the model version.
Mature GPT-4o capability set
Vision, function calling, streaming, fine-tuning, and predicted outputs cover many established production patterns.
Useful migration benchmark
The snapshot provides a fixed baseline for evaluating newer models against real historical GPT-4o workloads.
What to consider
Pinned behavior is intentionally static
The snapshot does not automatically inherit improvements introduced in newer models or model families.
128K context limit
The context window is much smaller than current models that support million-token input capacity.
Older knowledge cutoff
The documented GPT-4o knowledge cutoff is October 2023, so model knowledge should not be treated as current.
No native audio or video
The standard GPT-4o snapshot accepts text and image input; native audio and video require other model variants.
Keep the model version constant
Test the November 2024 GPT-4o snapshot on your own inputs
Run the same screenshots, documents, schemas, tool calls, and prompts against gpt-4o-2024-11-20 to create a reproducible baseline before changing prompts or migrating models.
Start FreeConnect your own OpenAI API key and evaluate the dated snapshot under the same conditions your production application uses.
Common Questions
What is GPT-4o-2024-11-20?
GPT-4o-2024-11-20 is a dated snapshot of OpenAI's GPT-4o model. It lets developers lock the November 2024 GPT-4o version for consistent behavior across tests and production deployments.
Why use GPT-4o (2024-11-20) instead of the GPT-4o alias?
Use the dated snapshot when model consistency matters. It helps keep the underlying GPT-4o version fixed while you change prompts, tools, schemas, image processing, or other application logic.
How much does GPT-4o (2024-11-20) cost?
The GPT-4o pricing profile is $2.50 per 1M standard input tokens, $1.25 per 1M cached input tokens, and $10.00 per 1M output tokens.
What is the context window of GPT-4o-2024-11-20?
The snapshot supports a 128,000-token context window and up to 16,384 output tokens.
Does GPT-4o-2024-11-20 support images?
Yes. The snapshot accepts text and image input and generates text output, making it useful for repeatable screenshot, document-image, chart, and other vision evaluations.
Does the GPT-4o November 2024 snapshot support function calling and structured outputs?
Yes. GPT-4o supports function calling and structured outputs, along with streaming, fine-tuning, and predicted outputs.
Is GPT-4o-2024-11-20 a reasoning model?
No. GPT-4o is a non-reasoning model and does not expose adjustable reasoning-effort controls.
Can GPT-4o-2024-11-20 be used as a migration baseline?
Yes. A dated snapshot is well suited to migration testing because it gives you a fixed GPT-4o reference for comparing vision accuracy, structured outputs, tool behavior, latency, and cost against a newer model.
Does GPT-4o-2024-11-20 support native audio?
No. The standard GPT-4o snapshot supports text and image input with text output. Native audio and realtime voice use separate OpenAI model variants.
Model information
Last updated
Specifications, pricing, modalities, capabilities, and snapshot information on this page are based on the official OpenAI GPT-4o model documentation. OpenAI lists gpt-4o-2024-11-20 as a GPT-4o snapshot that can be used to lock a specific model version for consistent behavior.