OpenAI API+3 Mixed

GILD Review

GILD's call and the buying facts behind it.

From GILD research

Should we adopt OpenAI API?

How much migration work will the OpenAI API's shutdowns and model retirements put on your roadmap?

30-Second TakePick the OpenAI API if GPT models belong inside your own software and you want each new model on release day.
Best fitTeams shipping OpenAI models inside their own products and internal tools, with engineers to own the integration and its evaluations.GILD Review
Consider another workflowIf the goal is coding help for your own engineers, adopt a finished agent such as Codex and skip building one on the API.GILD Review

Drawn from 154 sources and 9 operator insights

GILD analysis

OpenAI API Review

Build on the OpenAI API when GPT models belong inside your own software and you want each new model on release day. DoorDash and Grab keep Azure OpenAI as a second route behind their gateways, and Asana and Ploy cut agent costs on GPT-6.1 Sol and GPT-5.6 Sol by fixing caching first. Your team owns evaluations and migrations, with as little as 2 weeks' retirement notice for early-release models.

AI Model APIs
AI Costs

The decision

Best fit: Teams shipping OpenAI models inside their own products and internal tools, with engineers to own the integration and its evaluations.
Consider another workflow: If the goal is coding help for your own engineers, adopt a finished agent such as Codex and skip building one on the API.
Adoption condition: The team needs an evaluation set per task and a project hard spend limit before launch. Its usage tier must cover production traffic.

Basis: OpenAI documentation and pricing, the independent Artificial Analysis Intelligence Index, and 136 public accounts, including DoorDash's InfoQ talk and engineering write-ups from Asana, Grab, Ploy and Wealthsimple. They include measured benchmarks and evaluations such as Zep's long-context benchmark and You.com's 64-trial consistency study.

Buying & deployment factsWhat a team building on it should verify before production traffic grows.

How you buy it

US$2 in, US$10 out

Per million tokens for GPT-6.1 Sol. Luna is US$0.10 and US$0.50, Astra US$10 and US$50.

Source 1, OpenAI, Pricing: Pricing Source 2, OpenAI, Rate limits: Rate limits

Details

Usage based, per million tokens in US dollars on prepaid credits. For prompts up to 272,000 tokens, GPT-6 Luna US$0.10 input and US$0.50 output. GPT-6.1 Sol US$2 and US$10. GPT-6 Astra US$10 and US$50. Scale Tier and Reserved Tier sell more predictable capacity.

Cost drivers

Batch halves it

Batch and Flex halve prices. Cache reads are 5% of input on GPT-6.1 Sol and Terra, 10% on most others.

Source 1, OpenAI, Pricing: Pricing Source 5, developers.openai.com, GPT-6.1 Sol model page: GPT-6.1 Sol model page Source 6, developers.openai.com, API changelog: API changelog

Details

Model choice and output tokens, including reasoning, drive cost. Prompts over 272,000 input tokens cost 2x input and 1.5x output. Cache reads cost a tenth of the input rate on most models, or 5% on GPT-6.1 Sol and GPT-6.1 Terra. Writes cost 1.25x for the default 30-minute cache, or 2.5x for a 12-hour cache since October 20, 2026. Batch and Flex halve prices. Fast mode doubles them, and regional processing adds 10%.

Where it runs

US, EU, UAE

Regional processing options. Azure and Bedrock also run GPT-6 and bill on their own.

Source 3, OpenAI, Data controls: Data controls Source 7, azure.microsoft.com, Azure OpenAI pricing (Microsoft): Azure OpenAI pricing (Microsoft)Source 8, Amazon Web Services, Amazon Bedrock pricing (AWS): Amazon Bedrock pricing (AWS)

Details

OpenAI offers regional processing in the US, the EU, and the UAE, plus storage-only residency in seven more regions for eligible projects. Microsoft Azure and Amazon Bedrock also run GPT-6 models and bill through their own services.

What to verify

US$500 to US$200,000

The monthly usage cap, from the Build tier to Grow. Models have published shutdown dates.

Source 2, OpenAI, Rate limits: Rate limits Source 3, OpenAI, Data controls: Data controls Source 4, OpenAI, Deprecations: Deprecations

Details

Usage tiers cap monthly usage, from US$500 on Build to US$200,000 on Grow, and set per-model rate limits. Zero Data Retention eligibility depends on the endpoint. Batch and the Agents API keep application state. Models have published shutdown dates.