Loading page
OpenAI
GPT-3.5 Turbo
GPT-3.5 Turbo is a chat-oriented language model from OpenAI, positioned as an older, lower-cost tier in the company's GPT lineup for general text and conversational tasks.
Input from
$0.250 / 1M tokens
across 2 providers
API Pricing
Cheapest on OpenRouter — 67% below avg| Provider | Input / 1M | Output / 1M |
|---|---|---|
| $0.250 | $0.750 | |
| $0.500 | $1.50 | |
| $1.50 | $2.00 |
Prices updated daily. Last check: Sep 20, 2026
GPT-3.5 Turbo pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Intelligence
5.5 / 100
Coding
10.7 / 100
Reasoning & Knowledge
- MMLU-Pro46.2%
- GPQA Diamond29.7%
Math
- MATH-50044.1%
Benchmarks measured Sep 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- OpenAI
- Modalities
- Text
Capabilities
- Tool Calling
- No
- Open Source
- No
Strengths & Limitations
Strengths
- Long-established model with extensive public prompt patterns, tutorials, and community-tested integrations
- Uses OpenAI's standard chat completions message format, so code written for it ports easily to later GPT models
- Available in multiple dated snapshots, which lets teams pin a specific revision for reproducibility
- Generally low latency and high throughput relative to larger GPT-4 class models, suiting bulk text processing
- Well suited to bounded tasks like classification, tagging, summarization, and format conversion
- Broad third-party tooling and framework support due to years of ecosystem adoption
Limitations
- Superseded in capability by OpenAI's GPT-4 and GPT-4o generations, including their smaller variants
- Weaker at multi-step reasoning, complex code generation, and long-form instruction following than newer models
- Proprietary — served via API only, with no published weights for self-hosting
- Context window varies by dated snapshot and is small compared with modern long-context models
- Our database has no confirmed benchmark scores or throughput measurements for this entry
Key Features
•OpenAI chat completions API with system/user/assistant message roles
•Multiple pinned model snapshots for version stability
•Text generation, summarization, classification, and extraction
•Multilingual text handling
•Fine-tuning support offered by OpenAI for this model family
•Deprecation-aware versioning through OpenAI's model lifecycle policy
About GPT-3.5 Turbo
GPT-3.5 Turbo is an OpenAI model in the GPT series that predates the GPT-4 family. It was originally introduced as the model behind ChatGPT's free tier and became a widely adopted default for chat-completion style API workloads. Within OpenAI's catalog it sits below the GPT-4 and GPT-4o generations in capability, and it has largely been superseded by the smaller, cheaper members of those later families for new projects.
GPT-3.5 Turbo is accessed through OpenAI's chat completions interface and works with the same message-role structure (system, user, assistant) used by later GPT models, which makes prompt formats and client libraries broadly portable across the family. It handles general-purpose English and multilingual text generation, summarization, classification, extraction and light code assistance. We do not track a confirmed context window, modality list, or benchmark scores for this entry in our database, so buyers should check their provider's model card for the exact snapshot (GPT-3.5 Turbo has shipped in several dated revisions with differing context limits) before committing to it.
In practice GPT-3.5 Turbo is chosen for high-volume, latency-sensitive text pipelines where the task is well-specified and the marginal quality of a newer model is not required, or for legacy applications that were built and evaluated against it. Teams starting fresh usually compare it against the compact tiers of newer OpenAI generations, which typically offer stronger reasoning and instruction following at similar operating cost.
Common Use Cases
GPT-3.5 Turbo fits high-volume, cost-sensitive text workloads where the task is narrow and well-defined: intent classification, sentiment and topic tagging, short summarization, data cleanup and reformatting, template-driven copy, FAQ-style chat responses, and synthetic data generation for evaluation sets. It is also commonly retained in legacy production systems whose prompts and guardrails were tuned specifically against it, where swapping models would require re-validation. It is a weaker choice for agentic pipelines, difficult coding tasks, math-heavy reasoning, or long-document analysis; those workloads generally point toward the GPT-4o generation or a dedicated reasoning model.
Frequently Asked Questions
How much does GPT-3.5 Turbo cost to run?
Costs depend on which provider serves the model, whether you are billed for input or output tokens, and whether you use batch, cached, or fine-tuned variants. Because those rates change frequently, see the pricing table on this page for current per-provider numbers rather than relying on figures quoted elsewhere.
What is GPT-3.5 Turbo best used for?
Narrow, high-throughput text tasks — classification, tagging, extraction, short summaries, format conversion, and simple conversational replies. It is a reasonable default when the prompt is tightly specified and volume matters more than reasoning depth.
Should I use GPT-3.5 Turbo or a newer OpenAI model?
For new projects, most teams compare it against the compact tiers of OpenAI's GPT-4o generation, which generally handle reasoning, instruction following, and code better. GPT-3.5 Turbo remains relevant mainly for legacy deployments already tuned to it, or where a specific pinned snapshot is required for reproducibility.
Can GPT-3.5 Turbo process images?
We do not track confirmed image-input support for this model. Vision capability arrived in OpenAI's later GPT-4 class models, so if your workload requires image understanding, verify modality support on your provider's model card before building against GPT-3.5 Turbo.
How large is the GPT-3.5 Turbo context window?
Our database does not carry a confirmed context length for this entry. GPT-3.5 Turbo has shipped in several dated snapshots with different context limits, so check the specific revision your provider serves before designing long-prompt workflows.