VerdictPal
Join

Models / Other

Atria Dawn Preview

Atria · Released 2026-09-11

Shanghai AI Laboratory's MIT-licensed agentic preview, built on the 744B-parameter MoE GLM-5.2 foundation. Text-only. 256K context, 65,536 max output. Hosted at api.atria-asi.ai as Atria-Dawn-Preview; no public per-token list as of 20 Sep 2026. No AA Intelligence Index or Vals Index row as of 20 Sep 2026.

Open-weight research and engineering agent for plan, code, experiment, and retry loops — not a multimodal chat model.

Composite—
Context256K
Input / 1M—
Output / 1M—
Knowledge cutoffunpublished
Statussolid
Not enough families covered yet.Add more benchmark rows for Atria Dawn Preview to see the family profile.
Editor's note
Recorded from the Hugging Face card and the hosted API docs (256K context, 65,536 max output, text-only input, MIT weights, Chat Completions / Messages / Responses). The 14 Sep 2026 arXiv report (2609.15818) is the first full account after a quiet 11 Sep weight drop. Vendor BrowseComp, BFCL v4, CyberGym, and SWE-bench Pro tables stay off the ledger. No independent AA Intelligence Index or Vals Index row as of 20 Sep 2026.
Benchmark placements

Where Atria Dawn Preview places on each public benchmark source that publishes a row. Rank counts every model with a latest row in the same test — not a universal quality score.

No benchmark rows yet for Atria Dawn Preview.

Family context

The three highest-scoring models with pages in each capability family. Where Atria Dawn Preview shows up, it's highlighted.

Family coverage

A high composite that hides a weak family is a trap. These bars surface the families where this model hasn't been publicly tested, and where it leads.

No family coverage yet.
Price vs. performance
0255075100$0.10$0.30$1$3$10$30input USD / 1M tokens · log scale →← intelligence

Hover any dot for name, score, and input price. Keyboard: tab through the top twelve, or use the ranking below.

AnthropicOpenAIMistralAlibabaDeepSeekGoogleMoonshotMetaxAIMiniMaxZhipuXiaomiOtherMeituanCohereNVIDIA

Full benchmark ledger

Every catalog benchmark for Atria Dawn Preview. Scores link to the original source; gaps mean no public row exists yet.

0 sourced rows

0 of 37 catalog benchmarks have a sourced row for Atria Dawn Preview.

Changelog
  1. MIT BF16 and FP8 weights on Hugging Face and ModelScope; hosted API as Atria-Dawn-Preview.
  2. Technical report posted as arXiv:2609.15818.
Sources

Questions this model page answers

Every answer below is assembled from the dated fields on this page. Nothing is written separately for search.

What is Atria Dawn Preview?

Atria Dawn Preview is a Other model: Shanghai AI Laboratory's MIT-licensed agentic preview, built on the 744B-parameter MoE GLM-5.2 foundation. Text-only. 256K context, 65,536 max output. Hosted at api.atria-asi.ai as Atria-Dawn-Preview; no public per-token list as of 20 Sep 2026. No AA Intelligence Index or Vals Index row as of 20 Sep 2026.

What is Atria Dawn Preview for?

Open-weight research and engineering agent for plan, code, experiment, and retry loops — not a multimodal chat model.

What does Atria Dawn Preview cost?

Atria Dawn Preview lists — input and — output per 1M tokens. Pricing last checked 2026-09-20.

How large is the Atria Dawn Preview context window?

Atria Dawn Preview publishes a 256K token context window.

What is the VerdictPal status of Atria Dawn Preview?

Atria Dawn Preview carries Solid status on the model atlas.

The Pack · Editorial newsletter

New tools in your inbox. Free.

One short email when a tool ships or changes status. No tracking, no third-party analytics. Unsubscribe in one click.