OpenAI: GPT-3.5 Turbo Instruct

openai/gpt-3.5-turbo-instruct · OpenAIUnmeasurable

Tokenizer lineage (measured): OpenAI cl100k
Declared tokenizer tag: GPT
Entered catalogue: 28 September 2023
Last tested: 3 September 2026
Measurement: Unmeasurable: prompt caching corrupts its token accounting
Evidence confidence: Moderate

Why moderate: near-exact agreement (41 of 41 strings) with ibm-granite/granite-4.0-h-micro, short of a full exact signature.

Evidence

✓ near-exact agreement (41/41) with ibm-granite/granite-4.0-h-micro

Compare with IBM: Granite 4.0 Micro

Report a correction

Identity statement

OpenAI: GPT-3.5 Turbo Instruct is OpenAI's model, and its measured tokenizer sits in OpenAI's own signature group alongside its sibling models. No other maker's measured model shares it so far.

Whose vocabulary this is

Tokenizer: cl100k_base
Origin of tokenizer: OpenAI tiktoken, open source
Model maker: OpenAI

cl100k_base is an open vocabulary table anyone can build on. Sharing it says which dictionary the model reads with, and nothing about who made the model.

What we checked

Four different kinds of evidence, and what each one showed

We checked	What we found	Where it came from
How it counts tokens	Not measured yet	we measured it
How its API is set up	A combination of settings no other listing in the catalogue uses	api_647692d5
How much it can read and write	Reads up to 4,095 tokens at onceOpenRouter's listing shows 3,685 output tokens, which is 90% of the context window, the figure it carries where the provider declares no maximum.	its own listing
Who served our requests	OpenAI	we measured it, 3 September 2026

Closest measured models

Agreement across the mutually clean test strings

IBM: Granite 4.0 Microibm-granite/granite-4.0-h-micro41/41 · 100%IBM: Granite 4.1 8Bibm-granite/granite-4.1-8b41/41 · 100%Microsoft: Phi 4microsoft/phi-441/41 · 100%OpenAI: GPT-3.5 Turboopenai/gpt-3.5-turbo41/41 · 100%OpenAI: GPT-3.5 Turbo 16kopenai/gpt-3.5-turbo-16k41/41 · 100%OpenAI: GPT-4 Turboopenai/gpt-4-turbo41/41 · 100%

Click a row to open the full comparison.

The measured fingerprint

What each test string cost this model, in tokens

We sent OpenAI: GPT-3.5 Turbo Instruct fifty short pieces of text and recorded what each one cost it in tokens. The bars below are those costs. Two models built on the same tokenizer produce the same bars; a model built on a different one produces a different set, which is what makes this a fingerprint.

All 50 rows, 9 excluded as corrupt

English and whitespace

en-prose14

en-long16

spaces-203

spaces-603

tabs-203

newlines-204

mixed-ws5

Digits

digits-93

digits-124

digits-3010

digits-sep7

float-long9

CJK

zh-common18

zh-long25

zh-rarex

ja-kana14

ja-kanji17

kox

Other scripts

ru16

ar25

he24

hi32

th25

elx

Emoji

emoji-basicx

emoji-skinx

emoji-zwj-familyx

emoji-zwj-x3x

emoji-flags24

emoji-profx

Rare Unicode

math32

boxdrawx

combining14

cjk-ext-b13

surrogates40

zalgo36

rtl-mix11

Code

py-code25

py-indent15

json20

html16

regex44

camel5

snake6

Repetition and encodings

rare-word-x531

repeat-tok22

base6435

hex11

url17

uuid27

An x marks a row excluded as corrupt (caching or provider interference detected during measurement). Overhead subtracted: 8 prompt tokens.

Declared record

What the catalogue claims about this model

This model is a variant of GPT-3.5 Turbo tuned for instructional prompts and omitting chat-related optimizations. Training data: up to Sep 2021.

Modalities: text->text
Declared tokenizer: GPT
Prompt price: $1.50 / M tokens
Completion price: $2.00 / M tokens

Supported parameters

frequency_penaltylogit_biaslogprobsmax_tokenspresence_penaltyresponse_formatseedstopstructured_outputstemperaturetop_logprobstop_p

defaults: {}

Catalogue entry

History

Every observation, kept as taken

28 September 2023Enters the OpenRouter catalogue declared as GPT.
3 September 2026Fingerprinted in the YFarmX catalogue sweep · 41 of 50 strings measured clean.
3 September 2026YFarmX assessment: consistent with the OpenAI cl100k family, moderate confidence.