I agree with the commenters that your question is either ill-defined or nonsensical. It all hinges on what you mean by "tokens", and what it would mean to "not have tokens."
If you mean multi-character sequences that are "chunked" together into larger words, consider Brainfuck — or actually most esolangs, including my favorite, Befunge — where every "command" is a single character. However, on the other hand, many Brainfuck interpreters actually treat
[-] as a single "chunk" meaning "set the current tape cell to zero", so in that sense maybe you'd call
[-] a "token".
On the other hand, if you consider single characters to be "tokens", then maybe we need to go the other direction, and find programming languages in which everything kind of flows together without divisions. The first thing that pops to mind is Inform 7 (yes, that's Inform 7 source code!), where meaning is conveyed through English prose. Obviously Inform 7 has a parser that deals in words, but those words don't directly correspond to anything like
LET in Basic; it's more complicated than that.
Or, consider Piet, an esolang in which programs are expressed as two-dimensional pixel paintings. Does a pixel count as a "token"? Does a rectangle of pixels count as a "token"?
In the same vein — of trying to break away from the strictures of linear words and into more free-flowing worlds — consider noit o' mnain worb, a nondeterministic cellular automaton. It uses ASCII characters to represent cells in the automaton; but do those characters each count as a "token", or are the relevant building blocks of worb programs more like "corridors" and "gates"?
For that matter, consider Wireworld, a better-known cellular automaton. Does Wireworld count as a "language"? and if so, does "it" "have" "tokens"?
For that matter, consider the "language" in which ENIAC was programmed. It certainly didn't have "tokens"! But if you were to write down the plugboard configuration for a given "program", I suppose that you'd pick some way of writing it down that involved ASCII and a grammar of tokens. I don't guess you'd communicate the plugboard configuration with a pen-and-ink drawing. (But maybe you would! Who knows!)
For that matter, what would you say about plain old x86 machine code? That's a "computer language" if ever there was one... but it doesn't have "tokens", does it? All it has is a stream of bytes, one after the other, which are chunked together to form instructions. Or does an instruction (such as
C3) count as a "token"? Does it matter if the same set of bytes (or some of them, at least) could be reinterpreted by another part of the program as a piece of data, a jump address, et cetera?
Basically, you need to define what you mean by "language" and "token"; and once you do that, I think you'll find that your question answers itself.