GPT-3.5 Turbo · 1106
Pinned November 2023 snapshot of GPT-3.5 Turbo.
About GPT-3.5 Turbo · 1106
The 1106 checkpoint of GPT-3.5 Turbo arrived in November 2023 promising a dramatic speed leap — benchmarks confirmed it ran 6 to 11 times faster than its June 2023 predecessor, and at $0.50 per million input tokens it remained far cheaper than GPT-4 variants. For simple conversational tasks and rapid prototyping, that combination of speed and low cost still has genuine appeal. In practice, though, this particular checkpoint proved divisive. Developers praised the price and throughput but quickly ran into a notable behavioral regression: an unusually high rate of refusals ("I'm sorry, I can't do that") and broken function-calling that corrupted structured outputs, prompting some teams to roll back to earlier versions. It carries a September 2021 knowledge cutoff, a 16,385-token context window, and no vision capability. OpenAI itself now recommends GPT-4o mini as a more capable, cheaper, and multimodal replacement. This model suits cost-sensitive, non-agentic text tasks where occasional refusals are tolerable — but it is not the right choice for function-heavy or production-critical workflows.
Best for
- Low-cost conversational AI and chat interfaces where speed is the top priority
- Simple text generation, summarization, and analysis tasks that do not rely on structured output or function calling
- Rapid prototyping and proof-of-concept work before graduating to a more capable model
- Cost-sensitive applications with straightforward, non-critical text tasks
- High-throughput pipelines where per-token cost must stay low and occasional refusals can be filtered
Specs & capabilities
How GPT-3.5 Turbo · 1106 stacks up — intelligence, speed, context, and modalities.
Intelligence
Low
Speed
Slow
Context window
16,385 tokens
Max output
4,096 tokens
Knowledge cutoff
September 1, 2021
Supported endpoints
v1/chat/completions · v1/responses · v1/assistants · v1/batch · v1/fine-tuning
Input and output
Input: Text
Output: Text
Availability notes
Fine-tuning supported
Frequently asked questions
What does this model cost?
$0.50 per million input tokens and $1.50 per million output tokens.
How large is the context window?
16,385 tokens, with a maximum output of 4,096 tokens per response.
Can it process images or other file types?
No. GPT-3.5 Turbo 1106 is text-only — it cannot accept images, audio, or other non-text inputs.
What is it genuinely bad at?
Function calling and structured output. The 1106 checkpoint is widely reported to produce incomplete function arguments and degraded JSON output compared to earlier GPT-3.5 versions, making it unsuitable for agent or tool-use workflows.
How does it compare to GPT-4o mini?
OpenAI recommends GPT-4o mini over this model. GPT-4o mini is cheaper per token, supports vision, and avoids the refusal and function-calling regressions that affected the 1106 checkpoint.
What is the knowledge cutoff?
September 2021. It has no awareness of events, models, or developments after that date.