AionLabs: Aion-2.0 vs Mancer: Weaver (alpha)

AionLabs: Aion-2.0 is a text model from AionLabs, released on 2026-02-23. Mancer: Weaver (alpha) is a text model from Mancer, released on 2023-08-02. Both are served as text-in, text-out models through api14, so neither one accepts images or audio on input or produces anything other than text on output. The main structural difference between them is age and focus: Aion-2.0 is a newer general purpose model, while Weaver is an older alpha release that has been around long enough to be a known quantity for a particular style of prompting.

The two diverge most sharply on context and on what you can ask them to do. Aion-2.0 has a context window of 131072 tokens, while Weaver's is 8000 tokens, so Aion-2.0 is the one to reach for when you need to fit long documents, large code files or extended conversations into a single call. Aion-2.0 also supports reasoning and tool use, including the reasoning and include_reasoning parameters, tools and tool_choice, so it can call functions and expose a reasoning pass. Weaver supports none of those; instead it exposes a wide set of sampling controls such as frequency_penalty, presence_penalty, repetition_penalty, min_p, top_a, top_k, logit_bias, logprobs, top_logprobs, seed and stop. Both accept temperature, top_p, max_tokens and response_format, so structured output is available on either side.

For practical work, the choice usually comes down to task shape. If you are building an agent that needs to call tools, or you want the model to reason through a problem before answering, or your prompt plus retrieved context runs past a few thousand tokens, Aion-2.0 is the natural pick. If you are doing short-form generation where you want fine control over sampling, such as creative writing, character dialogue or experiments where you tune repetition and presence penalties and compare seeds, Weaver gives you the knobs for that. Weaver is also the cheaper of the two on both input and output, so for high volume, short prompts where its smaller context is not a constraint, it can be the more economical option. Aion-2.0 is the better fit when capability matters more than per call cost.

Calling either model through api14 works the same way, because the gateway exposes an OpenAI-compatible API. You point your client at the api14 base URL, send your request to the chat completions endpoint with your api14 key, and set the model field to either "aion-labs/aion-2.0" or "mancer/weaver". Everything else in the request body follows the OpenAI schema, so messages, temperature, max_tokens and response_format carry over unchanged. Switching between the two is a one line edit to the model id, though you should keep in mind that parameters are not perfectly interchangeable: tools, tool_choice and reasoning only apply to Aion-2.0, while the penalty, logprob and sampling parameters only apply to Weaver.

There are still cases where Weaver is the better pick even though Aion-2.0 is the more capable model overall. If your prompts are short, your budget is tight and you want to tune sampling behaviour precisely, Weaver's parameter set and lower price make it a reasonable default. It is also a sensible choice when you are reproducing work that was written against an older model and you want the same generation behaviour rather than a newer model's output. In those situations the smaller context window and the absence of tool use are acceptable trade-offs, and you can keep using the same api14 endpoint and the same request shape.