ComputeUnion

Changelog

We continuously fix price data errors, add new platforms, and document every change transparently. Price accuracy is the backbone of this project.

19
Total updates
4
Price fixes
1
New platforms
Datayunwu

Model Update: 5 new models on YunWu AI

YunWu AI added 5 new models (now 478 total).

Model Update: 13 new models on SambaNova

SambaNova added 13 new models (now 20 total).

Model Update: 10 new models on SambaNova

SambaNova added 10 new models (now 10 total).

Data302ai

Model Update: 9 new models on 302.AI

302.AI added 9 new models (now 16 total).

Model Update: 19 new models on Deep Infra

Deep Infra added 19 new models (now 31 total).

Model Update: 213 new models on OpenRouter

OpenRouter added 213 new models (now 312 total).

Datayunwu

Model Update: 276 new models on YunWu AI

YunWu AI added 276 new models (now 473 total).

Model Update: 86 new models on Together AI

Together AI added 86 new models (now 114 total).

Model Update: 6 new models on SambaNova

SambaNova added 6 new models (now 10 total).

Model Update: 7 new models on Mistral AI

Mistral AI added 7 new models (now 14 total).

Model Update: 106 new models on Together AI

Together AI added 106 new models (now 114 total).

New Platformqizhi

New platform: QiZhi AI

Added QiZhi AI (qizhi.lk666.ai), a Chinese AI API relay platform with 39 models tracked.

Merged H200/B200 fragmented GPU models

Genesis Cloud scraper lacked H200/B200 keyword mappings, creating new fragment models on every run. Prices were scattered and could not be aggregated. Fixed; H200 now aggregates 8 platforms, B200 aggregates 5 platforms.

Genesis Cloud upgraded to live price scraping

Genesis Cloud uses Webflow static rendering with pricing embedded in plain text. The old scraper searched for table rows and found nothing. Fixed with regex extraction; H100 $1.60/hr, H200 $2.80/hr, B200 $2.80/hr, RTX 4090 $0.55/hr now live.

Paperspace upgraded to live price scraping

Paperspace redesigned their pricing page to div cards. The old scraper searched for table rows and found nothing, falling back to static prices. Rewrote the parser; 11 GPU models (H100, A6000, V100, etc.) now update live.

Fixed Hyperstack broken scraper URL

Hyperstack moved their pricing page from /gpu-cloud-pricing to /pricing. The old URL returned 404, so the scraper was using stale 2024 static prices. Fixed URL and rewrote parser; prices are now live.

Fixed Datacrunch/Verda storing cluster prices as per-GPU

Verda pricing lists cluster prices (e.g. 8x B300 = $60/hr). The scraper stored the total cluster price instead of per-GPU price. B300 correct price is $7.50/hr but was stored as $60/hr (8x error). Fixed by detecting Nx prefix and dividing by N.

Fixed Shadeform price calculation (cents & multi-GPU)

Shadeform API returns hourly_price in cents for the entire machine. Old logic only divided by 100 when price >200, causing 199 cents to be stored as $199/hr. Fixed to always divide by 100 and by num_gpus.

Fixed Replicate multi-GPU cluster pricing

The Replicate pricing page lists both 1x and 8x GPU configs. The scraper stored the 8xH100 cluster price ($43.92/hr) as a single-GPU price. Fixed by filtering multi-GPU rows; H100 now correctly shows $5.49/hr.