What is AkashML
AkashML provides managed, low-latency inference for open models on the Akash Network through a playground and OpenAI- and Anthropic-compatible APIs.
Pay-as-you-go pricing
As of September 17, 2026, AkashML has no published monthly subscription minimum and charges credits by model usage. Purchased credits do not expire; trial credits are granted after payment-method verification and do expire. Users can configure manual purchases and automatic top-ups.
The live pricing table lists six models. Per million tokens, Qwen 3.8 27B costs US$0.25 input and US$2.20 output; Qwen 3.6 35B A3B costs US$0.10 and US$0.90; Llama 3.3 70B costs US$0.20 and US$0.52; GPT OSS 120B costs US$0.03 and US$0.17; GPT OSS 20B costs US$0.02 and US$0.10; and GLM 5.3 costs US$1.30 and US$4.40. Cached-input rates are published separately where supported.
Platform capabilities
The service includes per-model context and pricing metadata, API keys with request, token, concurrency and expiry controls, usage and cost dashboards, saved presets and a browser playground. The compatible endpoints support migration from common OpenAI and Anthropic SDK workflows.
Official sources: AkashML model pricing, AkashML billing and platform guide, AkashML models.