Abliterated Large Flash brings a million-token context to a lower price point. Keep room for long documents and conversation history, with input rates two-thirds lower and output rates 60% lower than Large v2.
Same context size. Lower token rates.
Flash and Large v2 both have a 1,048,576-token context window. Flash costs less per token going in and coming out, including discounted cached input.
| Tokens | Large v2 | Flash |
|---|---|---|
| Input | $3.00 | $1.00 |
| Cached input | $0.30 | $0.10 |
| Output | $5.00 | $2.00 |
- Lower input rate
- ⅔ less
- Lower output rate
- 60% less
Source: Abliteration pricing. Publisher rates exclude MiniRouter’s 5% fee. At those rates, Flash is $1.05 input, $0.105 cached input and $2.10 output per million tokens on MiniRouter. Your total depends on usage; see the model page for current rates.
Reason when you need to.
Unlike Large v2, Flash lets you turn reasoning off. Choose low, high or max when your task calls for it. High is the provider default.
Select an effort below to update the request. Set MINIROUTER_KEY to your API key, then copy the command into your terminal.
The provider default. Start here when your task needs reasoning.
curl https://api.minirouter.sh/v1/chat/completions \
-H "Authorization: Bearer $MINIROUTER_KEY" \
-H 'Content-Type: application/json' \
--data '{
"model": "abliteration/abliterated-model-large-flash",
"reasoning_effort": "high",
"max_tokens": 4096,
"messages": [
{
"role": "user",
"content": "Explain the tradeoffs of caching API responses for 60 seconds."
}
]
}'The example allows 4,096 output tokens, shared between reasoning and the answer. Changing the selector only edits the example; it does not send a request.
Make the switch explicit.
Use abliteration/abliterated-model-large-flash with your existing MiniRouter key. Chat Completions, Responses and Messages support streaming and non-streaming requests.
On MiniRouter, Flash accepts text and returns text, with up to 32,768 output tokens including reasoning. Keep the output cap explicit when comparing it with Large v2, and evaluate answer quality and token use on the tasks you actually run.
See the publisher’s model specifications and reasoning guide for upstream details.