← Back to Feed
retoor
retoor · Level 56859
fun

We live in the future!

AI pricing complex? Tiers? Different pricing on day and night according to China? No problem if you have Devii.

you> Please read all the fields and values on op my current page. As you can see, the money fields are empty. Please find in the openrouter docs what is exactly the price of this model and fill in the required fields..
  ↳ plan Read current page fields and fill pricing from OpenRouter do
  ✓ plan Read current page fields and fill pricing from OpenRouter do (success)
  ↳ get_page_context
  ✓ get_page_context (success)
  ↳ discover_elements 100
  ✓ discover_elements 100 (success)
  ↳ http_request GET https://openrouter.ai/api/v1/models
  ✓ http_request GET https://openrouter.ai/api/v1/models
  ! context compaction
  ! context compaction
  ! context compaction
  ! context compaction
  ↳ http_request GET https://openrouter.ai/api/v1/models/qwen/qwen3.7-flash/e
  ✓ http_request GET https://openrouter.ai/api/v1/models/qwen/qwen3.7-flash/e (success)
  ↳ read_element #gateway-model-form
  ✓ read_element #gateway-model-form (success)
  ↳ run_js const f=document.querySelector('#gateway-model-form'); const
  ✓ run_js const f=document.querySelector('#gateway-model-form'); const (success)
  ↳ run_js return document.querySelector('#gateway-model-form').innerHT
  ✓ run_js return document.querySelector('#gateway-model-form').innerHT (success)
  ↳ run_js return document.querySelector('#gateway-model-form').innerHT
  ✓ run_js return document.querySelector('#gateway-model-form').innerHT (success)
  ↳ fill_field e89
  ✓ fill_field e89 (success)
  ↳ fill_field e90
  ✓ fill_field e90 (success)
  ↳ fill_field e91
  ✓ fill_field e91 (success)
  ↳ fill_field e92
  ✓ fill_field e92 (success)
  ↳ fill_field e83
  ✓ fill_field e83 (success)
  ↳ discover_elements Tier
  ✓ discover_elements Tier (success)
  ↳ fill_field e1
  ✓ fill_field e1 (success)
  ↳ fill_field e2
  ✓ fill_field e2 (success)
  ↳ fill_field e3
  ✓ fill_field e3 (success)
  ↳ fill_field e4
  ✓ fill_field e4 (success)
Filled from OpenRouter qwen/qwen3.7-flash endpoints API (Alibaba base endpoint, context 1,000,000):

Price cache-hit / 1M: 0.0 -> 0.006 (from $0.000000006/token x 1M)
Price cache-miss / 1M: 0.0 -> 0.03 (from $0.00000003/token x 1M)
Price output / 1M: 0.0 -> 0.13 (from $0.00000013/token x 1M)
Price input / 1M: 0.0 -> 0.03
Context window: 0 -> 1000000
Tier-2 threshold: 0 -> 32000 (OpenRouter override min_prompt_tokens 32000: $0.10 in / $0.40 out / $0.02 cached)
Tier-2 cache-hit: empty -> 0.02
Tier-2 cache-miss: empty -> 0.10
Tier-2 output: empty -> 0.40
Not saved yet - review and press Save.

Only sad that I can't configure what tier to use :P
@admins If you never did, check the provider page for once and see the possibilities of this system. You'll see the lessons learnt and the effort for cheap but quality AI (transformers).

Should i automate this prompt?
Log in to vote

Comments

retoor retoor 
Price cache-hit / 1M: 0.0 -> 0.006 (from $0.000000006/token x 1M)

Sending fuck you to this model will cost you 0.000000018. For Devii it's two times that. For Dolphin like ten times that or something. As i said, chat is nothing. The automation makes it expensive.

retoor retoor

@dolphin how does this feel? You're cheap dude:

you> Please check the input fields on my page and update them with the correct information based on openrouter documentation regarding prices... The flash 2.5 model is what we need prices for to fill in.
Checking your form fields and pulling the 2.5 Flash price.

Checking the form fields and fetching prices.

You are on the dolphin edit page, but I'll check its fields and apply the 2.5 Flash pricing.

Filling the price fields with the official 2.5 Flash rates.

Fields updated from OpenRouter docs for google/gemini-2.5-flash (standard Google AI Studio endpoint, context 1,048,576):

Context window: 0 -> 1048576
Price cache-hit / 1M: 0.0 -> 0.03 (0.00000003 x 1M)
Price cache-miss / 1M: 0.0 -> 0.3 (0.0000003 x 1M)
Price output / 1M: 0.0 -> 2.5 (0.0000025 x 1M)
Source: OpenRouter endpoint pricing - prompt 0.0000003, completion 0.0000025, input_cache_read 0.00000003.

I did not save the form. Click Save to apply.
retoor retoor

It's not even the cheapest model but the jailbreak works specifically on this version.

retoor retoor

Can take some time before fake fish responds. It has a circuit breaker. A circuit breaker prevents DDoS'ing of the server when something goes wrong in any way. It's also the technique I've applied on all AI providers now to prevent getting blocked again. Our system kept on going hardcore with requests while that thing was warning for a ban.

retoor retoor ↳ @retoor

That's how we do. Hmm, 30 seconds? Should be configurable. I was thinking about an hour since we have fallback models. So it would go for an hour on fallback (mostly pricier or less quality, since the original model is always best task for the job and that's why it is configured...)

Screenshot 2026-11-14 06.46.44.png
retoor retoor ↳ @retoor

Actually, this can be a provider error - it worked for an hour or so. Or there were no requests in an hour and this is the first one. I should add extra information. Now I only know the remote model name. Not my local one. I white label them often. It's functional: we can make models that do not exist and thus give models who do not have vision, vision through a different model. And yes, financially it is just perfectly tracked. When configuring a vision model, you select one of devplace pre-configured and those should be financially configured.