Atria Dawn Preview scores 96.0% on the vendor's DeepSearchQA research benchmark and runs locally via vLLM 0.23 or newer, ready since its September 2026 launch. It suits engineering and research teams that want an open, self-hostable agent model, not shops needing vision or audio input.
An agentic foundation model from Shanghai AI Lab, Atria Dawn Preview scores 59.6% on SWE-bench Pro since its September 2026 release. Built through post-training on Zhipu AI's GLM-5.2 base rather than trained from scratch, it targets long-horizon research and engineering work over short chat replies.
Provider: Shanghai AI Lab · Family: Atria Dawn
More about Shanghai AI Lab on HokAI
Context window: 256,000 tokens
Input modalities: text, tool-calls · Output: text, tool-calls
About Atria Dawn Preview
Atria Dawn Preview is a 744 billion parameter mixture of experts model Shanghai AI Lab published on Hugging Face on September 11, 2026, with an FP8-quantized checkpoint following on September 12 and a formal announcement on September 15. It is not trained from scratch: the model is built through agentic post-training on top of Zhipu AI's GLM-5.2 base, the same foundation model Z.ai shipped in June 2026. The release came from ATRIA, a cross-institutional research initiative Shanghai AI Lab runs with Fudan University's Natural Language Processing Lab, led by Fudan associate professor Tao Gui, with Shanghai AI Lab director Bowen Zhou among the roughly 140 listed co-authors on the accompanying arXiv paper, Atria Dawn: The Dawn of Agentic Superintelligence. It sits at the start of what the paper frames as a new Atria Dawn model family, positioned as Shanghai AI Lab's first dedicated agentic foundation model rather than an extension of its existing InternLM chat model line.
The vendor's own benchmark table spans research, coding, tool use, document work and cybersecurity. It reports 96.0% on DeepSearchQA and 92.5% on BrowseComp, both research and web-browsing evaluations, 86.2% on MLE-bench Lite and 59.6% on SWE-bench Pro for coding and machine learning engineering, 77.0% on BFCL v4 and 53.8% on AutomationBench for tool use, 65.0% on Workspace-Bench and 50.3% on JobBench for document and delivery tasks, and 86.5% on CyberGym for vulnerability analysis. The paper claims the model is competitive with frontier agents and reaches the highest reported score on five of sixteen benchmarks, without naming which five. Independent verification is thin: aiweekly.co's review found that BenchLM, a third-party tracker, covers only 14 of the roughly 435 benchmarks relevant to a model like this, so most of the vendor's own numbers have not yet been checked by an outside evaluator.
Atria Dawn Preview ships with a 256,000 token context window, enough to hold a research paper, a codebase excerpt, or a long multi-step agent transcript in a single call. No independent long-context recall evaluation, such as a needle-in-haystack test, has been published for the model, so how well it holds up near the top of that window is unverified as of this writing.
The model card lists text-only input and output: there is no vision, audio, or video support documented, which sets it apart from the multimodal flagships most Western labs now ship by default. Its strength instead is tool use and agentic loop execution, reflected in the BFCL v4 function-calling benchmark and the AutomationBench score, and the vendor organizes its intended use around four areas it calls Discovery (research and evidence retrieval), Creation (software and application development), Delivery (document and data transformation) and Cybersecurity (vulnerability analysis and validation). Two agent harnesses, Codex and Claude Code, are listed as compatible integrations, meaning developers can point an existing agentic coding tool at Atria Dawn Preview instead of a proprietary model.
There is no published per-token pricing. The model ships as open weights under the MIT license, so the direct cost is whatever GPU infrastructure you provide to self-host it, not a metered API bill. Two hosted API endpoints went live alongside the release, an international one at api.atria-asi.ai and a China endpoint at discovery.intern-ai.org.cn, both OpenAI-compatible in format, but neither had a public price list as of the September 2026 launch.
Self-hosting requires SGLang version 0.5.13.post1 or newer, or vLLM version 0.23.0 or newer; both are recent releases, so older inference stacks will not load the checkpoint. Weights are published in BF16 and F32 for the base model, plus a separate FP8-quantized checkpoint aimed at lower-VRAM deployments, on both Hugging Face and ModelScope. No managed cloud listing on AWS Bedrock, Google Vertex AI, or Azure was found.
No system card, red-teaming partner list, or refusal-rate benchmark specific to Atria Dawn Preview has surfaced yet. That is a gap relative to the lab's own public safety work elsewhere: its Intern-SafeWork stack and its published value-alignment and adversarial-robustness benchmarks cover the general InternLM line, and director Bowen Zhou chairs the AI safety working group of China's national cybersecurity standards committee, TC260, but none of that has yet been tied specifically to this model's release documentation.
Teams running long, multi-step agent loops such as literature review into experiment design, coding, or vulnerability analysis get a free, self-hostable model built specifically for that workflow rather than adapted from a chat model. Teams that need multimodal input, a managed API with a published SLA and price list, or benchmark scores independently confirmed beyond the vendor's own paper should look at a competitor such as DeepSeek V4 or Qwen3-Max, or wait for third-party evaluation of Atria Dawn Preview to catch up.
The paper describes training through what it calls a Verifiable Experience Pipeline, connecting tool-mediated agent interactions to executable environments rather than relying only on static text corpora, and reports that human raters judged roughly one-third of completed AI-assisted tasks as infeasible without the model's help. Beyond that description, Shanghai AI Lab has not disclosed a training data cutoff date, an API input retention policy, or a stance on training future models on user inputs, and no SOC 2, HIPAA, or GDPR compliance statement was found for either hosted endpoint.
Pricing
Open-weight release under the MIT license; no published token pricing for the hosted API. Self-host via vLLM or SGLang, or use the early hosted endpoints at api.atria-asi.ai (international) and discovery.intern-ai.org.cn (China), neither of which documents a price as of launch.
Key Features
- Agentic Post-Training on a Frontier Base: Adds agentic post-training on top of a large third-party mixture-of-experts foundation model rather than training from scratch.
- Long-Context Window for Agent Runs: Wide enough to hold a full research paper, codebase excerpt, or multi-step agent transcript in a single call.
- Four Agentic Focus Areas: Purpose-built workflows for Discovery (research), Creation (coding), Delivery (document and report generation) and Cybersecurity (vulnerability analysis).
- MIT-Licensed Open Weights: Both the base and an FP8-quantized checkpoint are downloadable from Hugging Face and ModelScope for commercial self-hosting.
- Dual-Region Hosted API: A separate international endpoint and a China endpoint both went live at launch, though neither publishes per-token pricing yet.
Pros
- Ships as fully open, MIT-licensed weights, so teams can self-host, fine-tune, or run air-gapped without a vendor contract.
- A wide context window and agentic post-training suit long research and coding sessions better than a short-context chat model.
- Free to run at whatever infrastructure cost you provide, with no per-token vendor markup on top.
Cons
- No independently verified benchmark scores at launch: aiweekly.co reported that third-party BenchLM tracking covers only 14 of the roughly 435 relevant benchmarks.
- Text-only: no vision, audio, or video input, unlike multimodal frontier models from OpenAI, Google or Anthropic.
- No managed hosted pricing, SLA, or enterprise support published as of the September 2026 launch, so production reliability depends on self-hosting.
Benchmarks
- SWE-bench Pro: 59.6% vendor-reported · 16 Sep 2026 — Harder, longer real-repository coding tasks, % solved.
A benchmark is an exam, not the job. Scores transfer unevenly between tasks, so weigh the one closest to your workload and read every figure with its source.
Frequently Asked Questions
What does Atria Dawn Preview cost to run?
Atria Dawn Preview ships as free, open weights under the MIT license, with no published per-token price for the hosted API as of its September 2026 launch. Running it yourself means GPU infrastructure costs instead: a mixture-of-experts model this size needs multi-GPU serving even at FP8 precision.
How does Atria Dawn Preview compare to other coding models on benchmarks?
The vendor reports its strongest results on research-focused evals like DeepSearchQA and BrowseComp, ahead of its coding-specific scores, and its own paper claims a leading result on several of its evaluation suite, though it declines to specify which ones. Independent verification is limited, so treat the comparison as vendor-reported until a third party re-runs the numbers.
Is Atria Dawn Preview open source?
Yes. Shanghai AI Lab released the base and FP8-quantized weights on Hugging Face and ModelScope under the MIT license, which permits commercial use and modification. Local deployment is supported through recent releases of vLLM or SGLang.
What is Atria Dawn Preview's data handling policy for API users?
Neither Shanghai AI Lab nor the ATRIA project has said whether inputs sent to the hosted API get used for future training or how long they are kept, and no trust page or terms of service covering it was found as of September 2026. Anyone with sensitive data is better off self-hosting the open weights until that documentation appears.
Who should use Atria Dawn Preview, and who should avoid it?
It fits teams running long, multi-step agentic workflows: literature-driven experiment design, code implementation, or vulnerability analysis, the four areas the vendor built it for. Teams needing image or audio input, a managed API with published pricing and SLAs, or independently confirmed benchmark scores should pick a different model instead.