Models & prices.

Anthropic's model for demanding reasoning and long-horizon agentic work, served by Anthropic at 40% below its list price. Thinking is always on and billed as output. Anthropic's OpenAI-compatible API, which serves it, does not return the thinking and offers neither prompt caching nor response_format.

Anthropic's Claude Opus 5 (a legacy model; Anthropic's current Opus is 5.5), served by Anthropic at 40% below its list price. Adaptive thinking is on by default and billed as output. Anthropic's OpenAI-compatible API, which serves it, does not return the thinking and offers neither prompt caching nor response_format.

Anthropic's Claude Sonnet 5 (a legacy model; Anthropic's current Sonnet is 5.5), served by Anthropic at 40% below its list price. Adaptive thinking is on by default and billed as output; non-default temperature or top_p values are rejected. Anthropic's OpenAI-compatible API, which serves it, does not return the thinking and offers neither prompt caching nor response_format.

OpenAI's GPT-6 Astra reasoning model, served by OpenAI at 40% below its list price. A request whose prompt exceeds 272,000 tokens is billed at 2x input and 1.5x output prices in full, as OpenAI bills it. Prompt caching is off through Avenro, so every input token is billed at the input price. reasoning_effort accepts low, medium, high, xhigh and max.

• Prompts over 272,000 tokens: the whole request at $12.00 input and $45.00 output per 1M.

OpenAI's proprietary GPT-4.1 mini with a 1M-token context window, served through OpenAI at 40% below OpenAI's list price.

OpenAI's proprietary GPT-4o mini, served through OpenAI at 40% below OpenAI's list price.

Google's Gemini 3.8 Flash, served by the Gemini API's paid tier (which does not use prompts to improve Google's products) at 40% below its list price. Thinking tokens are billed as output. Google raises this price on 2027-01-01; Avenro withdraws the model at that moment until it posts the new price.

• Listed at this price until Jan 1, 2027, 12:00 AM UTC, when the provider's price changes.

xAI's Grok 4.6 reasoning model, served by xAI at 40% below its list price. A request whose prompt reaches 200,000 tokens is billed at double prices in full, as xAI bills it. Reasoning tokens are billed as output.

• Prompts of 200,000 tokens or more: the whole request at $2.40 input, $0.60 cached input and $7.20 output per 1M.

DeepSeek-V4-Pro (0813, open weights under the MIT licence), served by DeepSeek's own API, whose servers are in China. Avenro charges 40% below DeepSeek's peak-hour list price at all hours (DeepSeek halves its own price off-peak). Thinking mode is enabled with "thinking": {"type": "enabled"}; its reasoning arrives in reasoning_content and is billed as output.