GPT-4 was introduced on March 14, 2023 as a large multimodal model accepting text and image input while producing text. “Multimodal” in that historical description did not mean every developer received image access immediately or that the model could natively generate video. Release marketing, research capabilities and deployed API features should be audited independently.
Historical research cutoff: 25 October 2026. Primary-source historical claims are date-scoped; cloud entitlements, region-specific availability, current pricing and rate limits are not implied.
Release identity at a glance
- Provider
- OpenAI
- Model family
- GPT-4
- Public launch
- 2023-03-14
- API ID / checkpoint
- gpt-4
- API retirement status
- Not confirmed as retired at the research cutoff
The release and why it mattered
OpenAI’s technical report and announcement presented improvements in academic-style examinations, instruction following and factuality relative to its earlier systems. Those are vendor-reported research results under particular protocols, not proof that GPT-4 is error-free. The launch initially used phased access, and its status in the consumer ChatGPT product followed a distinct chronology from the developer API.
Capabilities and genuine use cases
The new practical opportunity was combining language reasoning with visual material, such as interpreting diagrams or discussing photographed text. Developers could also build drafting, analysis and coding workflows with text-only inputs. A visible demonstration should never be generalized to all images: small print, ambiguous charts and complex spatial detail require specific testing and human verification.
Integration and version identity
Preserve an exact dated snapshot when trying to reproduce output. `gpt-4` is a family-facing API alias and must not be treated as synonymous with every GPT-4 Turbo or GPT-4o revision. Separate system instructions, prompt examples, tool definitions and output schema in migration logs. For regulated or source-dependent applications, ground generated statements against original documents rather than asking the model to certify its own answers.
Limits and evidence gaps
The model research itself acknowledged hallucinations, reasoning failures and reliance on training-data patterns. Comparisons to later models are vulnerable to changed benchmark prompts, tool use and evaluator settings. Older public statements about ChatGPT Plus or API waitlists do not establish access for every region, subscription or cloud reseller on October 25, 2026.
How to choose or migrate
GPT-4 remains historically important for understanding the move from text completion to general-purpose vision-language systems. Use a currently supported model when building fresh production integrations and test the specific capability needed; do not substitute a later product’s media-generation or realtime features into the March 2023 GPT-4 record.
Reproducible evaluation checklist
Build three distinct test sets: the regular path, an adversarially ambiguous path, and a regression set drawn from failures. Hold prompts and tool access constant while comparing snapshots. Never use a consumer product announcement to fill an API region or account-entitlement field.
If the historical route is retired, keep the example as an archived request only. For active migration, identify a current model from the provider catalogue and perform a staging replay before changing production configuration.
Sources & Further Reading
- openai.com — documentation/reference (supplier-provided reference; not independently checked in this pass)
- openai.com — documentation/reference (supplier-provided reference; not independently checked in this pass)
- developers.openai.com — documentation/reference (supplier-provided reference; not independently checked in this pass)
Source documentation establishes vendor-published facts, not independently measured model quality, tenant access, contractual rights or current prices. Use a dated pricing ledger and an account-specific entitlement check for deployment decisions.