Azure OpenAI alternatives compared by price

The OpenAI API is the developer side of ChatGPT's maker, priced per million tokens with cached input at a tenth of the standard input rate. Batch and Flex processing cost half, Fast mode costs double, and the flagship has been repriced with almost every model generation since GPT-3 charged per thousand tokens.

OpenAI API starts at $0.10 per million input tokens, $0.5 per million output for gpt-6-luna (USD per million tokens, standard processing), goes up to $10 per million input tokens, $50 per million output for gpt-6-astra. Latest change: Cheapest model, output from $1.20 to $0.50 (-58%), Oct 2026.

Amazon Bedrock is where AWS customers buy Claude, Amazon Nova and other models without a separate vendor contract. Claude's top tier on Bedrock went from $8 and $24 per million tokens for Claude 2.1 to $15 and $75 for Opus 4, and now tracks Anthropic's own API exactly at $10 and $50 for Fable 5.1.

Amazon Bedrock starts at $0.30 per million input tokens, $2.5 per million output for Amazon Nova 2 Lite (USD per million tokens, on-demand in US East (N. Virginia), global cross-region rate where offered), goes up to $10 per million input tokens, $50 per million output, global for Claude Fable 5.1. Latest change: Claude flagship, output from $25 to $50 (+100%), Sep 2026.

Vertex AI is the Google Cloud route to Gemini, priced the same as Google's developer API but billed through a Cloud account. Gemini Pro went from $1.25 to $2 per million input tokens with Gemini 3, and the Flash tier has climbed from $0.15 to $0.75.

Vertex AI starts at $0.75 per million input tokens, $3.75 per million output, introductory until 2027 for Gemini 3.8 Flash (USD per million tokens, standard tier, prompts up to 200K tokens, global endpoint), goes up to $2 per million input tokens, $12 per million output for Gemini 3.1 Pro Preview. Latest change: Gemini Flash tier, output from $3 to $3.75 (+25%), Oct 2026.

TypeSafe AI sells Jev, a model that answers typed questions about program state with structured, calibrated decisions rather than free text, aimed at automation inside software. There are no plans or seats: usage is metered at $42 per billion input tokens and output is not charged, which the company positions against the per-token prices of general purpose language models.

TypeSafe AI starts at $0.042 per million input tokens, output tokens free for Jev (USD per million input tokens; output tokens are free).

The Claude API is how developers pay for Anthropic's models: per million tokens, with input and output priced separately and every model tier a fixed step apart. Since Opus 4.5 in late 2025 the top tier has dropped from $15 to $5 per million input tokens while a new tier above it, Fable 5.1, took the $10 slot.

Claude API starts at $1 per million input tokens, $5 per million output for Haiku 4.5 (USD per million tokens, standard processing), goes up to $10 per million input tokens, $50 per million output for Fable 5.1. Latest change: Sonnet tier, output from $15 to $10 (-33%), Jul 2026.

The Gemini API is Google's per token offer to developers and the only one of the big three with a free tier on most models. Pro models charge more above 200,000 tokens of prompt, and the newest Flash models are sold at an introductory rate that Google has already announced will double in 2027.

Gemini API starts at $0.25 per million input tokens, $1.5 per million output for Gemini 3.1 Flash-Lite (USD per million tokens, paid tier, prompts up to 200K tokens), goes up to $2 per million input tokens, $12 per million output for prompts up to 200K tokens for Gemini 3.1 Pro Preview, has a free plan.

The xAI API sells Grok models per million tokens with the same input and output split as its rivals, cached input at a discount and a doubled rate for long prompts. Its flagship sits between OpenAI's mid and top tiers on price.

xAI API starts at $1 per million input tokens, $2 per million output for grok-build-0.1 (USD per million tokens, prompts under 200K tokens), goes up to $2 per million input tokens, $6 per million output for grok-4.6.

Mistral's La Plateforme prices its models far below the US labs per million tokens, and unusually its flagship Large 3 costs less than the mid tier Medium 3.5. Open weight models like Ministral sit at the bottom with the same price for input and output.

Mistral API starts at $0.15 per million input tokens, $0.6 per million output for Mistral Small 4 (USD per million tokens), goes up to $1.50 per million input tokens, $7.5 per million output for Mistral Medium 3.5.

Perplexity's Sonar API is a web-grounded answer engine for developers. Token prices look like everyone else's, but each request also carries a fee that depends on how much web context is pulled in, which makes it a hybrid of per token and per call pricing.

Perplexity Sonar API starts at $1 per million input tokens, $1 per million output, plus $5 to $12 per 1,000 requests for Sonar (USD per million tokens, before per request fees), goes up to $3 per million input tokens, $15 per million output, plus $6 to $14 per 1,000 requests for Sonar Pro.

DeepSeek sells open-weight-lineage models through a plain per-token API at some of the lowest rates in the category. Peak input starts at $0.30 per million tokens and halves off peak.

DeepSeek API starts at $0.30 per million input tokens at peak, $1.20 per million output, half price off peak for deepseek-flash (USD per million tokens, peak rate, cache miss), goes up to $1.32 per million input tokens at peak, $3.96 per million output, half price off peak for deepseek-v4-pro.