Grok Composer 2.5 Fast: the 0.05x model and what it gives up

The lowest multiplier is not the lowest price

grok-composer-2.5-fast is billed at 0.05x. Nothing else in the catalog goes that low — multipliers here run from 0.05x up to 1x, with 0.1x covering most mainstream ids. Reading that number alone, it looks like the cheap option by a wide margin.

The multiplier is a discount, and a discount is only as good as the rate it applies to. Composer's official rate is $3 per million input tokens and $15 per million output, the highest official input rate of the three Grok text models. At 0.05x that becomes $0.15 per million input and $0.75 per million output.

Compare that with gemini-3-flash, which carries a plain 0.1x multiplier on a $0.50/$3 official rate and lands at $0.050 and $0.30 per million. Composer costs three times as much on input and 2.5 times as much on output. The id with the biggest discount and the id with the smallest bill are not the same id.

Against the other Grok ids, it depends on your output ratio

Within the Grok text line the picture flips depending on the shape of your traffic. grok-4.6 and grok-4.5 both cost $0.20 per million input and $0.60 per million output. Composer is cheaper on input ($0.15) and dearer on output ($0.75).

That gives a clean break-even: Composer is the cheaper of the two whenever your output is less than one third of your input, and the more expensive one above that. A 100,000-token prompt returning 2,000 tokens costs $0.0165 on Composer against $0.0212 on grok-4.6. Flip it to a 5,000-token prompt returning 8,000 tokens and it is $0.0067 against $0.0058, the other way round.

So the question is not which id is cheap, it is what your input-to-output ratio actually is. Pull that ratio off a day of real traffic before deciding; a review job that reads a large diff and answers in a paragraph sits on one side of the line, and a generation job that writes long files from a short brief sits on the other. One caveat on the arithmetic: the $0.20/$0.60 figures for grok-4.6 and grok-4.5 are the sub-200K tier, and xAI doubles both at or above 200,000 tokens in a single prompt — which is also the point where Composer's own window runs out.

curl https://token-share.app/v1/chat/completions \
  -H "Authorization: Bearer $TOKEN_SHARE_KEY" \
  -H "content-type: application/json" \
  -d '{
    "model": "grok-composer-2.5-fast",
    "max_tokens": 2048,
    "messages": [
      {"role": "user", "content": "Apply this patch and report any conflicts."}
    ]
  }'

What the id gives up on specifications

Composer lists text and tool-calling. It does not list thinking, which the other two Grok text ids do. If your prompts depend on an extended reasoning step, this is a hard difference rather than a quality judgement — the capability is simply not on the id.

The window is smaller as well: 200,000 tokens of context against 500,000 on grok-4.6 and grok-4.5, and a 32,768-token output ceiling against 65,536. A prompt that fits on 4.6 may not fit here, and a response that fits there may be cut in half.

None of that makes it the wrong choice. It makes it a choice with a shape: high input volume, short answers, no reasoning step, prompts that stay under 200,000 tokens. Traffic that matches gets a genuinely low bill; traffic that does not will either fail to fit or pay $0.75 per million on output for the privilege.

Common questions