Home / Add-ons / Procurement Report
Procurement Report — 11 AI API vendors with dated, sourced prices
A dated price-and-terms snapshot: 22 model rows across 11 vendors, 10 with published token prices, and a vendor-owned source URL for every number.
On this page · 12 sections
Who it is for
Procurement, finance and platform teams who need a defensible list price before a negotiation or a business case.
The problem it removes
- Vendor comparison blogs quote prices with no date and no source.
- A quote arrives and nobody knows whether it is above or below list.
- Residency and retention claims are repeated from memory.
What you get
- 11 vendors, 22 model rows, 10 with published prices
- Input, output and cached price per 1M tokens
- Residency, retention and no-training statements where published
- 44 source records, 30 unique vendor URLs, all dated 22 November 2026
- Quote-only rows are marked, never estimated
How to use it
1. Download
No account required. The ZIP contains both CSVs, the PDF, the README and the licence.
2. Shortlist
Filter the CSV by vendor, price tier or published residency.
3. Cite it
Every row carries the vendor URL and the date checked, so the number survives a review.
curl -sS https://plugsky.com/downloads/plugsky-procurement-2026-11.zip -o plugsky-procurement-2026-11.zip
What is in the report
The report is a dated snapshot, not a live feed. It answers one question well: what does a vendor publish as its list price, and what does it say about residency, retention and training, on 22 November 2026.
| File | Contents | Use |
|---|---|---|
| report.pdf | 4 pages: method, price table, terms table, limits | Attach to a business case |
| procurement.csv | 22 model rows, 19 columns | Filter, sort, join to your own usage |
| sources.csv | 44 source records, 30 unique URLs | Check any claim without trusting us |
| README.md | Column definitions and method | Hand the dataset to an analyst |
Coverage
Eleven vendors, two models sampled per vendor where prices are published. The range below is for the sampled models only, not the full catalogue — a vendor's cheapest and most expensive model can differ by 80×.
| Vendor | Sampled models | Input range | Output range | EU residency published |
|---|---|---|---|---|
| OpenAI | 2 | $0.10–$8.00 | $0.50–$40.00 | Yes |
| Anthropic | 2 | $0.10–$8.00 | $0.50–$40.00 | No (US option) |
| 2 | $0.30–$2.00 | $2.50–$10.00 | Not stated on the pricing page | |
| Mistral AI | 2 | $0.15–$0.70 | $0.60–$2.10 | Yes |
| Groq | 2 | $0.075–$0.15 | $0.30–$0.60 | No (US storage) |
| xAI | 2 | $1.25–$2.00 | $2.50–$6.00 | US option; no EU endpoint published |
| Together AI | 2 | $0.30–$1.40 | $1.20–$4.40 | Not published |
| Fireworks AI | 2 | $0.30–$2.00 | $1.20–$10.00 | US option; no EU endpoint published |
| DeepInfra | 2 | $0.06–$1.30 | $0.18–$2.60 | Not published |
| OpenRouter | 2 | $0.10–$2.00 | $0.50–$10.00 | Not published |
| Cohere | 1 | quote-only; no published token price on the page checked | Not published | |
What a row looks like
Each CSV row carries the commercial fact and the evidence for it. A sample row, shortened:
vendor: OpenAI
model_name: gpt-6.2-astra
input_usd_per_1m: 8.00
output_usd_per_1m: 40.00
cached_input_usd_per_1m: 0.40
pricing_status: published
enterprise_pricing: Published list price; enterprise/committed terms not published
data_residency: US and EU (EEA+Switzerland) regional processing
retention: Abuse-monitoring logs up to 30 days; ZDR on approval
training_on_api_data: No, API data not used to train since 2023-03-01
price_source_url: https://platform.openai.com/docs/pricing
date_checked: 2026-11-22
The four source columns are the part that makes the row auditable. A reader can open the price page, the training statement, the retention article and the residency page independently, without trusting a summary.
How the data was collected
Every numeric cell is either a vendor-published value or the literal string not published. Nothing is inferred from a blog, a memory or a reseller's summary. Each row carries four source columns — price, training, retention and residency — plus the date checked. The sources file lists every URL and its HTTP status at collection time.
Quote-only is a first-class value. Two rows are explicitly quote-only — Cohere Command and one Groq enterprise row — and enterprise or committed terms are never estimated. A procurement table that invents a discount is worse than one with a gap, because the gap is visible.
Using MCP? The free/paid tool map is in the MCP documentation.
The honest gaps
- EU residency is the weakest field. Only OpenAI and Mistral publish a first-party EU inference endpoint in the sampled set. Anthropic, Fireworks and xAI publish a US option; Groq publishes US-only storage; Google, Together, DeepInfra and OpenRouter publish no regional option on the pages checked.
- Cloud resale channels are excluded. Bedrock, Azure/Foundry and Vertex price pages are region-and-model calculators. A single number would be a guess, so they are out.
- List prices only. Committed spend, volume tiers and private offers are not public and are not in the dataset.
- No latency or quality. This is a price-and-terms product. The separate Benchmark Report covers measured latency on the Plugsky catalogue.
Using MCP? The free/paid tool map is in the MCP documentation.
A dated change to watch
One row is a trap if you read it without the note: Google's Gemini 4 Carbon list price doubles on 1 December 2026, from $2 / $10 to $4 / $20 per 1M tokens. The CSV records today's price and the change in the notes; the report does not pretend the December number is current. That is what a dated snapshot is for — it tells you what was true when it was checked, and what is already scheduled to change.
Who should skip it
If you already have a procurement team with a contract database, this adds little. If you need negotiated pricing, it cannot help. And if you only need one vendor's current price, the vendor's own page is free — this product is the dated, source-linked comparison across eleven of them.
Version and licence
Version 2026-11. Five files, about 37 KB uncompressed. Commercial use inside your organisation is covered, including procurement documents and board papers. Redistribution or resale of the dataset is not. The licence is in the ZIP.
Questions
What exactly do I get?
A ZIP with report.pdf (4 pages), procurement.csv (22 rows), sources.csv (44 records), README.md and LICENSE.txt.
Are the prices negotiated or list?
List prices only. Enterprise and committed terms are quote-only and marked as such, never estimated.
Which vendors are covered?
Eleven, including OpenAI, Anthropic, Google, Mistral, Cohere, Groq, Together, Fireworks, DeepInfra, OpenRouter and xAI.
Does it cover EU residency?
Only where the vendor publishes it. OpenAI and Mistral publish a first-party EU inference endpoint; most others do not, and the report says so.
How current is it?
Every row is dated 22 November 2026. Prices change; the report is a snapshot, not a feed.
Sources
- OpenAI API pricing — checked 22 November 2026
- Anthropic pricing — checked 22 November 2026
- Groq models and pricing — checked 22 November 2026
- Plugsky pricing and fair-use limits — checked 22 November 2026
Related add-ons
Benchmark Report
A dated PDF and CSV of p50, p95 and p99 latency for all 39 public model IDs, measured by a five-minute production probe from 16 to 22 November 2026.
AI Cost Index Pro
A dated JSON dataset of Plugsky plan prices, intervals and the live model capability matrix, built for procurement comparison and internal budgeting.
Free download. No account required. The ZIP includes the PDF, the data and the licence.
Download the ZIPCompare add-ons