LL Alpha Free, a free model with no named maker, has appeared in OpenCode's /models picker. OpenCode's data page for ll-alpha lists context, output and release as "Unknown" while showing 28B tokens and 1.9K unique users on Oct 16. The OpenCode models page lists it as free. According to @notjazii, it had been out since the previous night and OpenCode hadn't announced it on X; "stealth" is that poster's word, and OpenCode's pages don't use it.

Our take: an unannounced free model that drew 28B tokens in a day is worth watching, but until someone names its maker it isn't a clear place to send private code.

The poster's screenshot shows the picker filtered to "free", with LL Alpha Free at the top under "OpenCode". Beam Free and LongCat 2.5 Preview Free each appear twice, under both "OpenCode" and "OpenCode Go", while LL Alpha Free appears only once in the visible rows.

OpenCode model picker filtered to free, with LL Alpha Free selected above Nemotron, Beam, LongCat, Space Bunny, MiMo and DeepSeek entries
OpenCode model picker filtered to free, with LL Alpha Free selected above Nemotron, Beam, LongCat, Space Bunny, MiMo and DeepSeek entries

Image credit: @notjazii on X

The technical receipt

The entries read:

LL Alpha Free | ll-alpha-free | https://opencode.ai/inference/openai/v1/chat/completions | @ai-sdk/openai-compatible
LL Alpha Free | Free | Free | Free | -
ll-alpha: Context Unknown · Output Unknown · Release Unknown

The data page sits under the path /data/unknown/. The pricing row shows free input, output and cached reads, with no end date or limit listed. The endpoint and package are shared by many other models on the same page, including GLM 5.3 Flash, Kimi K3, DeepSeek V4.1 Flash, MiniMax M3 and Big Pickle, so they say nothing about who built it.

For scale, the same pricing table lists the paid DeepSeek V4.1 Flash at $0.30 input, $1.20 output and $0.006 cached read, and a separate DeepSeek V4.1 Flash Free row at Free. That is a reference for what "free" replaces on this page, not a sign that LL Alpha Free is a DeepSeek model.

Who makes it, and is it GLM?

The poster's hunch is that it is probably a GLM model, based on how it reasons. Nothing in OpenCode's records supports that, and the same models page lists GLM 5.3 Flash and "GLM-5.3-Flash Free" under their own names. The poster's second screenshot shows the model replying "My developer isn't disclosed, so I can't share that information" when asked who made it, and then "I can't share who developed or trained me; that isn't disclosed."

OpenCode session where LL Alpha Free replies that its developer isn't disclosed when asked which lab made it
OpenCode session where LL Alpha Free replies that its developer isn't disclosed when asked which lab made it

Image credit: @notjazii on X

Day-one usage

The data page's only dated point is Oct 16: 28B tokens, 1.9K unique users, 40,153 completed sessions, rank #31, $0.00 total spend and a 91% cache ratio on input tokens. It lists average tokens per session as 696K.

The page's own figures give about 697K tokens per session (28B ÷ 40,153) and roughly 14.7M tokens per user (28B ÷ 1.9K), so the average checks out. On the page's peer ranking by recent token volume, LL Alpha Free sits at #31, level with longcat-2.0 (28B). It is below qwen3.7-plus (61B), glm-5.2 (55B), kimi-k3 (45B) and mimo-v2.5-pro (35B), and ahead of jev-1.13 (24B) and qwen3.8-max (17B). Our read: free access inflates usage, and this is a single day of data, so it shows curiosity more than staying power.

Sources