Curiosity · AI Model

GPT-4 Turbo

GPT-4 Turbo was OpenAI's mid-cycle upgrade to the original GPT-4, shipping with a 128K-token context window, function calling, JSON-mode outputs, and optional vision via gpt-4-turbo-vision. It drove the first wave of serious enterprise GPT-4 deployments in 2024 and is still reachable via the legacy endpoints today.

Model specs

Vendor
OpenAI
Family
GPT-4
Released
2023-11
Context window
128,000 tokens
Modalities
text, vision
Input price
$10/M tok
Output price
$30/M tok
Pricing as of
2026-04-20

Strengths

  • Reliable JSON-mode structured outputs
  • Mature ecosystem — most tools support gpt-4-turbo out of the box
  • Broadly available via Azure OpenAI with strict SLA

Limitations

  • Outperformed by GPT-4o on latency, multimodal, and price
  • No native audio, weaker vision than gpt-4o
  • OpenAI considers it legacy — new features target the -o and -5 families

Use cases

  • Legacy enterprise chatbots built on the GPT-4 era
  • Long-document RAG pipelines using 128K context
  • Structured-output generation via JSON mode
  • Function-calling agent loops from the 2024 stack

Benchmarks

BenchmarkScoreAs of
MMLU≈86%2024-01
HumanEval≈87%2024-01
MATH≈52%2024-01

Frequently asked questions

What is GPT-4 Turbo?

GPT-4 Turbo is OpenAI's November 2023 upgrade to GPT-4, offering a 128K-token context window, JSON-mode structured output, function calling, and optional vision input at a meaningfully lower price than the original GPT-4 model.

Is GPT-4 Turbo still supported?

Yes — OpenAI continues to host GPT-4 Turbo endpoints for enterprise customers who depend on its behavior, though newer workloads should target GPT-4o or GPT-5 for better latency and pricing.

What is the context window of GPT-4 Turbo?

GPT-4 Turbo supports a 128,000-token context window with up to 4,096 output tokens per completion.

Should I migrate off GPT-4 Turbo?

Most new projects should target GPT-4o or GPT-5 mini for equal or better capability at lower cost and latency. Migrate when you can re-test prompts and structured outputs.

Sources

  1. OpenAI — GPT-4 and GPT-4 Turbo — accessed 2026-04-20
  2. OpenAI — DevDay 2023 announcement — accessed 2026-04-20