Aster Inference

Aster Inference is a commercial OpenAI-compatible API from Aster that serves open-weight models including gpt-oss-120b variants, GLM 5.2 and Kimi K3 ¹. Developers can point OpenAI-style clients at Aster's base URL and use common SDKs and agent frameworks rather than adopting a proprietary client layer ¹². Aster positions the service as an output of its autonomous-research work on inference optimization; its published throughput comparisons are first-party benchmarks, not independent performance results ¹³.

Pricing is usage-based and model-specific. Current published examples include gpt-oss-120b-fast at $0.30 per million input tokens and $1.20 per million output tokens, with higher rates for larger/newer model surfaces such as Kimi K3 ¹. Access is self-serve through an API key and prepaid credits, with volume pricing available by contact ¹. There is no stable free tier in the current canonical product record, so value depends primarily on whether Aster's latency/throughput advantages matter for the workload.

Aster states that prompts and outputs are processed in memory and are not stored by default; it retains token counts for billing ¹. That is a useful privacy default for an inference provider, though AiToolMap found no broad independent security-assurance corpus. Y Combinator corroborates Aster as an active Spring 2026 company ⁴, and APIs.io independently maps the same OpenAI-compatible endpoint and developer surface ². An independent provider integration reports live compatibility testing across Aster models, including successful streaming/function-calling paths plus concrete quirks such as strict-schema failures for some gpt-oss requests and serialized parallel tool calls ⁵. This is more decision-useful than generic directory coverage, but it remains a small technical sample rather than a large reliability benchmark.

Aster Inference is best suited to developers already using OpenAI-compatible tooling who want to swap in hosted open-weight models with minimal integration work ¹². Agent and coding workflows benefit from the broad client compatibility and the availability of reasoning/tool-capable models. Teams needing contractual enterprise controls, a long production track record or a large independent SLA/reliability corpus should evaluate those requirements separately rather than inferring them from benchmark speed claims.