Mistral: Mistral Large 4

mistralai/mistral-large-4-0 · Mistral AINot yet established

Declared family: Mistral
Declared tokenizer tag: Mistral
Entered catalogue: 6 October 2026
Last tested: 8 October 2026
Measured signature: tk_e242e95c

Identity statement

Mistral: Mistral Large 4 has a clean measured fingerprint that currently sits alone: no other measured model shares its signature.

Verification note

Checked against the primary artefacts, 8 October 2026

Measured on 8 October 2026, two days after it was listed, and measured again the same day pinned to Mistral's own endpoint: both readings agree on all 50 strings. Mistral Large 4 runs a tokenizer no other model on record uses, Mistral's own included. It agrees with Mistral Large 2512 and Mistral Medium 3.5 on 13 of the 50 strings, and its nearest match anywhere is OpenAI's vocabulary at 31. The new vocabulary is cheaper on exactly the text the older one was dear on: a 12-digit number costs 4 tokens where Mistral Large 2512 charges 12, a line of emoji 10 where it charges 20, JSON 20 where it charges 25 and a Chinese sentence 12 where it charges 16, while English prose and Python cost the same.

Mistral Large 4 on OpenRouter

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	Counted cleanly, and no other model in the catalogue counts the same way	tk_e242e95c
How its API is set up	A combination of settings no other listing in the catalogue uses	api_c182db23
How much it can read and write	Reads up to 1,048,576 tokens at once, writes up to 262,144	its own listing
Whether it thinks before answering	Reasons when asked to, at high, none, high by default	its own listing
Who served our requests	Mistral	we measured it, 8 October 2026

Closest measured models

Agreement across the mutually clean test strings

OpenAI: GPT-5 Nanoopenai/gpt-5-nano31/50 · 62%OpenAI: GPT-5.5 Proopenai/gpt-5.5-pro31/50 · 62%OpenAI: gpt-oss-20bopenai/gpt-oss-20b31/50 · 62%OpenAI: gpt-oss-120bopenai/gpt-oss-120b31/50 · 62%OpenAI: gpt-oss-safeguard-20bopenai/gpt-oss-safeguard-20b31/50 · 62%OpenAI: GPT-4.1 Nanoopenai/gpt-4.1-nano31/50 · 62%

Click a row to open the full comparison.

The measured fingerprint

What each test string cost this model, in tokens

We sent Mistral: Mistral Large 4 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

English and whitespace

en-prose14

en-long18

spaces-203

spaces-603

tabs-205

newlines-204

mixed-ws7

Digits

digits-93

digits-124

digits-3010

digits-sep7

float-long9

CJK

zh-common12

zh-long16

zh-rare19

ja-kana11

ja-kanji13

ko11

Other scripts

ru11

ar9

he9

hi11

th18

el14

Emoji

emoji-basic10

emoji-skin20

emoji-zwj-family11

emoji-zwj-x333

emoji-flags16

emoji-prof21

Rare Unicode

math27

boxdraw16

combining11

cjk-ext-b13

surrogates41

zalgo34

rtl-mix5

Code

py-code25

py-indent15

json20

html16

regex43

camel6

snake6

Repetition and encodings

rare-word-x536

repeat-tok22

base6433

hex11

url17

uuid27

Overhead subtracted: 3 prompt tokens.

Declared record

What the catalogue claims about this model

Mistral Large 4 is a frontier multimodal (text and image input) model from Mistral AI built for reasoning, coding, and agentic workloads. It offers a 512K-token context window with up...

Modalities: text+image->text
Declared tokenizer: Mistral
Prompt price: $0.68 / M tokens
Completion price: $2.09 / M tokens

Supported parameters

frequency_penaltyinclude_reasoninglogprobsmax_tokenspresence_penaltyreasoningreasoning_effortresponse_formatseedstopstructured_outputstemperaturetool_choicetoolstop_logprobstop_p

defaults: {}

Catalogue entry

History

Every observation, kept as taken

6 October 2026Enters the OpenRouter catalogue declared as Mistral.
8 October 2026Fingerprinted in the YFarmX catalogue sweep · 50 of 50 strings measured clean.
8 October 2026Fingerprinted again, to check the reading held · 50 of 50 strings measured clean.