OpenAI dated snapshot
GPT-5 (2025-08-07)
The version-pinned August 7, 2025 snapshot of GPT-5, designed for teams that needed stable model behavior across coding, reasoning, and agentic production workloads.
- Context window
- 400K
- tokens
- Max output
- 128K
- tokens
- Input
- $1.25
- per 1M tokens
- Cached input
- $0.125
- per 1M tokens
- Output
- $10.00
- per 1M tokens
01 / Overview
What GPT-5 (2025-08-07) Is
GPT-5 (2025-08-07) is the dated API snapshot that pins GPT-5 to its August 7, 2025 release version, giving applications a fixed model identifier for consistent behavior instead of following an alias that can change over time.
A Reproducible GPT-5 Release, Not a Separate Model Family
The identifier gpt-5-2025-08-07 represents a snapshot of GPT-5 rather than a different capability tier.
OpenAI uses snapshots to let developers lock an application to a specific model version so that performance and behavior remain consistent. This matters when prompts, evaluations, structured outputs, tool policies, or business logic have been validated against one exact release.
The snapshot corresponds to the original GPT-5 API generation launched on August 7, 2025. That release focused heavily on coding and agentic tasks, including complex codebase work, detailed instruction following, long-running tool use, configurable reasoning, response verbosity, and custom tools.
A dated snapshot creates a clean evaluation boundary. Teams can distinguish changes caused by their application from changes caused by a newer model revision.
That stability has a lifecycle cost. OpenAI deprecated this snapshot and scheduled its removal from the API for December 11, 2026. The recommended replacement is gpt-5.6-sol.
- Fixed August 7, 2025 version of GPT-5.
- Designed for reproducible production behavior.
- Useful for regression tests and historical evaluations.
- Preserves the original GPT-5 coding and agentic capability profile.
- Deprecated with a defined migration deadline.
- Provider
- OpenAI
- Model family
- GPT-5
- Snapshot
- 2025-08-07
- API ID
- gpt-5-2025-08-07
- Status
- Deprecated
- Knowledge cutoff
- Sep 30, 2024
- Input modalities
- Text, Image
02 / Use cases
Where a Dated GPT-5 Snapshot Matters
The strongest reason to retain gpt-5-2025-08-07 is reproducibility: it gives teams a stable reference for production behavior, regression testing, incident analysis, and migration evaluation.
Version Pinning Turns Model Behavior into a Testable Dependency
An alias is convenient when an application should follow a provider's current model revision. A snapshot is useful when the exact model version is part of the application's tested configuration.
That distinction matters for production AI systems. A prompt can produce different wording, tool choices, reasoning depth, or structured values after a model revision even when the application code is unchanged.
Pinning gpt-5-2025-08-07 gave teams a stable target while they validated prompts, agent loops, tool schemas, and quality thresholds.
The same snapshot becomes especially valuable during migration. A team can replay a representative evaluation set against both the old snapshot and its replacement, identify behavioral differences, and decide whether prompt or application changes are necessary.
After deprecation, the snapshot's primary value shifts from new deployment to controlled migration and historical comparison.
- 01
Production pinning
Keep model behavior tied to a known release while prompts, schemas, and application logic are validated.
- 02
Regression testing
Replay the same evaluation set against a fixed GPT-5 version to detect application or model-behavior changes.
- 03
Migration comparison
Compare the dated snapshot with gpt-5.6-sol on the same prompts before changing the production model identifier.
- 04
Historical baseline
Preserve a reference point for understanding how an application behaved on the original August 2025 GPT-5 release.
03 / Pricing
GPT-5 (2025-08-07) Pricing
The GPT-5 snapshot uses the GPT-5 pricing profile: $1.25 per million input tokens, $0.125 per million cached input tokens, and $10.00 per million output tokens.
Preserve Cost Alongside Quality in Migration Evals
A migration test should compare more than answer text.
The same workload can produce different token usage when reasoning behavior, response length, or agent execution changes. A replacement model may use a different number of reasoning tokens, tool calls, retries, or output tokens even if it ultimately solves the same task.
That makes the old snapshot's pricing useful as a baseline. Record the complete cost of representative GPT-5 workflows before moving them.
Cached input is priced at one tenth of standard input, which can reduce repeated-prefix costs for stable system instructions, tool definitions, or repository guidance.
For agentic tasks, evaluate cost per completed workflow rather than one API turn.
1M tokens · USD
- Input
- $1.25
- Cached input
- $0.125
- Output
- $10.00
Example: 10K input + 2K output
- Input cost
- $0.0125
- Output cost
- $0.0200
- Estimated total
- $0.0325
04 / Context
400K Context in a Version-Pinned Model
GPT-5 (2025-08-07) provides a 400,000-token context window and supports up to 128,000 output tokens, matching the original GPT-5 model profile while keeping the underlying version fixed.
Context Strategy Should Stay Constant During Model Comparison
The snapshot can hold substantial working sets: source files, requirements, retrieved documents, tool definitions, prior messages, and intermediate agent results.
For coding and agentic workloads, that capacity made GPT-5 suitable for tasks that needed more than an isolated prompt or code snippet.
When evaluating a migration, context strategy should be controlled carefully. If the old snapshot receives a curated set of files while the replacement receives a different retrieval set, the comparison no longer isolates the model change.
The same applies to conversation truncation, RAG settings, system prompts, and tool definitions.
A strong migration evaluation first holds context constant across models. Context strategy can then be optimized separately after behavioral differences are understood.
Context window
400,000
Max output
128,000
For a clean migration test, replay the same system prompt, history strategy, retrieved context, tool definitions, and user input against the snapshot and replacement model.
05 / Reasoning
Minimal to High Reasoning as a Migration Variable
The August 2025 GPT-5 release supports minimal, low, medium, and high reasoning effort, with medium documented as the launch default.
Preserve Reasoning Effort Before Comparing Models
GPT-5 introduced the minimal reasoning setting to reduce thinking time when a request did not need extensive deliberation.
low, medium, and high provided progressively more reasoning effort, with higher settings oriented toward quality and lower settings toward speed.
For snapshot testing, reasoning effort is part of the baseline. Comparing GPT-5 at high against a replacement at a low-effort setting can produce misleading conclusions about quality, latency, and cost.
The first migration pass should record the existing GPT-5 reasoning configuration and choose an intentional equivalent or candidate setting on the replacement model.
Reproduce current behavior
Record the reasoning effort used by the existing GPT-5 integration and replay representative workloads without changing unrelated variables.
Optimize after migration
Once quality parity is understood, test alternative reasoning settings on the replacement model for better latency or cost.
06 / Capabilities
The Original GPT-5 Developer Feature Set
GPT-5 (2025-08-07) captures the original GPT-5 developer generation with streaming, function calling, structured outputs, configurable verbosity, custom tools, parallel tool calling, and agentic coding behavior.
Snapshot Stability Includes More Than Text Quality
Model migrations can affect tool behavior as much as natural-language output.
The original GPT-5 release emphasized reliable multi-step tool execution. OpenAI described the model as better at following tool instructions, handling errors, chaining many calls, and using sequential or parallel actions without losing track of the overall task.
GPT-5 also introduced a verbosity parameter with low, medium, and high settings. This gave applications direct control over the model's default answer length.
Custom tools allowed developer-defined tools to accept free-form plaintext instead of requiring JSON-only arguments. Grammar constraints could be applied when a particular free-form syntax was required.
The official GPT-5 profile lists streaming, function calling, and structured outputs as supported, with text and image input and text output.
These details should be part of migration testing. Evaluate not only final answer quality but also tool selection, argument generation, progress behavior, and schema reliability.
- Supported
Structured outputs
Use schema-constrained responses in applications that require machine-readable output.
- Supported
Function calling
Connect the fixed GPT-5 snapshot to application-defined actions and agent workflows.
- Supported
Custom tools
Use free-form tool payloads for code, SQL, shell-like text, and other developer-defined formats.
- Supported
Verbosity
Control default answer detail with low, medium, or high verbosity.
- Supported
Image input
Include visual context alongside text in coding, reasoning, and agent tasks.
- Not listed
Fine-tuning
The GPT-5 model profile lists fine-tuning as unsupported.
07 / Evaluation
Strengths and Limitations
The value of gpt-5-2025-08-07 is predictability rather than novelty: it provides a fixed historical GPT-5 baseline for production systems, but its approaching shutdown means every remaining integration needs a deliberate migration plan.
Snapshot strengths
Version-pinned behavior
A dated identifier locks the application to a specific GPT-5 release instead of relying on an alias that may move to another version.
Reproducible evaluations
Teams can rerun the same prompts and agent scenarios against a stable model target for regression and migration testing.
Original GPT-5 agent capabilities
The snapshot preserves the August 2025 generation built around coding, tool orchestration, custom tools, and configurable reasoning.
Clear migration baseline
A fixed old version makes behavioral differences easier to attribute when testing gpt-5.6-sol or another replacement.
What to consider
Scheduled shutdown
OpenAI will remove gpt-5-2025-08-07 from the API on December 11, 2026.
Deprecated snapshot
The snapshot is no longer a suitable starting point for new production integrations.
Older model generation
Newer GPT models provide newer reasoning behavior, model knowledge, context capabilities, and tool ecosystems.
Migration work is unavoidable
Version pinning protects short-term consistency but does not remove provider lifecycle deadlines; applications must validate and adopt a replacement.
Build a migration baseline
Compare your pinned GPT-5 snapshot before shutdown
Replay representative production prompts against gpt-5-2025-08-07 and newer models, then compare response quality, reasoning behavior, token usage, cost, and regressions in EidoStack.
Start FreeThe gpt-5-2025-08-07 snapshot is scheduled for API shutdown on December 11, 2026. OpenAI recommends gpt-5.6-sol as the replacement.
Common Questions
What is GPT-5 (2025-08-07)?
GPT-5 (2025-08-07) is the dated snapshot gpt-5-2025-08-07. It pins GPT-5 to its August 7, 2025 release version so applications can depend on a specific model revision.
How is gpt-5-2025-08-07 different from gpt-5?
gpt-5-2025-08-07 is a dated snapshot, while gpt-5 is the general model identifier. OpenAI describes snapshots as a way to lock in a specific model version so performance and behavior remain consistent.
Is gpt-5-2025-08-07 deprecated?
Yes. OpenAI has deprecated gpt-5-2025-08-07.
When will gpt-5-2025-08-07 shut down?
OpenAI's deprecation schedule lists December 11, 2026 as the API shutdown date for gpt-5-2025-08-07.
What should replace gpt-5-2025-08-07?
OpenAI's official deprecation schedule recommends gpt-5.6-sol as the replacement.
Why use a dated model snapshot?
A dated snapshot helps keep model performance and behavior consistent across production requests, evaluations, regression tests, and incident analysis.
How much does gpt-5-2025-08-07 cost?
The GPT-5 pricing profile lists $1.25 per 1M input tokens, $0.125 per 1M cached input tokens, and $10.00 per 1M output tokens.
What is the context window of gpt-5-2025-08-07?
The GPT-5 snapshot has a 400,000-token context window and supports up to 128,000 output tokens.
What reasoning levels are available?
The original GPT-5 API generation supports minimal, low, medium, and high reasoning effort. OpenAI documented medium as the launch default.
What is the knowledge cutoff?
The official GPT-5 model profile lists September 30, 2024 as the knowledge cutoff.
Does the GPT-5 snapshot support image input?
Yes. GPT-5 accepts text and image input and produces text output.
Does GPT-5 support structured outputs and function calling?
Yes. The official GPT-5 model profile lists both structured outputs and function calling as supported.
What should be tested when migrating from this snapshot?
Replay representative production prompts with the same context strategy and compare final quality, structured outputs, tool choices, reasoning configuration, token usage, latency, retries, and total workflow cost.
Why keep this model page after the snapshot shuts down?
The page remains useful as historical documentation for migration research, regression analysis, old production configurations, pricing comparisons, and understanding the original GPT-5 API generation.
Model information
Last updated
The snapshot identity, specifications, pricing, capabilities, and lifecycle information on this page are based on OpenAI's official GPT-5 model documentation, the August 7, 2025 GPT-5 developer launch, and OpenAI's API deprecation schedule.