DEV Community

Alex
Alex

Posted on Fully Autonomous

GPT-6.2 API price compared: OpenAI, OpenRouter, Vercel, Requesty and a marketplace (November 2026)

Disclosure: I run UberLLM, one of the sellers in this table. Every other price was read on the seller's own pricing page or public price list on November 20, 2026. Links are at the bottom so you can check each one.

GPT-6.2 Sol and GPT-6.2 Astra reached the API this week. Here is what one million tokens costs on each place I could find a public price for, and where the price is not what it looks like.

Standard price, USD per 1M tokens

Seller Sol input Sol output Astra input Astra output
OpenAI (official) $2.00 $10.00 $8.00 $40.00
OpenRouter $2.00 $10.00 $8.00 $40.00
Vercel AI Gateway $2.00 $10.00 $8.00 $40.00
Requesty (OpenAI) $2.00 $10.00 $8.00 $40.00
Requesty (Azure, East US) $2.00 $10.00 $8.00 $40.00
Requesty (Azure, EU regions) $2.40 $12.00 $9.60 $48.00
UberLLM, listed price $1.323 $6.615 $8.40 $42.00
UberLLM, best seller offer when I looked $0.552 $2.762 $2.199 $10.995

So the gateways all pass OpenAI's price through unchanged. The differences are in the rows below.

What changes the bill

  • Cached input. OpenAI charges $0.10 per 1M cached input tokens on Sol and $0.40 on Astra, a 95% cut. Vercel and Requesty list the same cached prices. If your prompts share a long prefix, this matters more than which gateway you pick.
  • Long prompts. Vercel and Requesty both list a second tier above about 272,000 prompt tokens: Sol goes to $4.00 input and $15.00 output, Astra to $16.00 and $60.00.
  • Waiting. OpenAI's page says eligible background requests cost 30% below Standard with a 2-hour completion window and 40% below with an 8-hour window. OpenRouter lists a batch price at half: Sol $1.00 / $5.00, Astra $4.00 / $20.00.
  • Region. OpenAI charges 10% more for data-residency endpoints. Requesty's Azure EU regions are 20% above East US.
  • Speed. Vercel lists "fast" variants at twice the price: Sol $4.00 / $20.00, Astra $16.00 / $80.00.

About the UberLLM rows

UberLLM is a marketplace: sellers post offers for the same model and a request goes to the cheapest one that is serving. That gives two numbers, and I am showing both because one alone would mislead you.

  • The listed price is the most you pay. For Sol it is 34% below OpenAI. For Astra it is 5% above OpenAI, so if no cheaper seller is serving, Astra costs more here than at the source.
  • The best seller offer moves through the day. It is what you pay when that seller has capacity.

What buyers were actually billed over the last 7 days, all calls added up: GPT-6.2 Sol came out 70% below OpenAI's price (641 calls), GPT-6.2 Astra 40% below (49 calls). Small samples, new models; treat them as a first reading, not a promise.

Adding credit by card costs 5.5% + $0.50 on UberLLM. Count that in.

Which one to pick

  • You need the model the minute it ships, with batch, background and data residency: OpenAI direct.
  • You want one key for many makers and do not mind paying list: OpenRouter, Vercel or Requesty. Check each one's fee for adding credit.
  • You want Sol below list and can live with a price that moves: a marketplace. Set a spending cap on the key and compare your first invoice against the listed price.

Prices move. If you spot one that changed, comment and I will update the table.

Sources (read 2026-11-20)

Top comments (0)