Baidu lists ERNIE 5.5 Flash Preview with a 1-million-token context

Baidu has added ERNIE 5.5 Flash Preview to its Qianfan model list, giving developers access to an endpoint with a one-million-token context window. The listing was updated on 10 November 2026 and identifies the endpoint as `ernie-5.5-flash-preview`. ¹

The same entry lists a maximum input of one million tokens and a maximum output of one million tokens. Baidu’s default service limits are shown as 60 requests per minute and one million tokens per minute. Those figures describe the published Qianfan defaults, not a guarantee that every account or deployment will receive identical throughput.

What Baidu has listed

ERNIE 5.5 Flash Preview appears in the ERNIE section of Qianfan’s text-generation model list. The page also places the model in its deep-thinking catalogue, using the same endpoint identifier and one-million-token limits. ¹ The preview label matters: this is a model exposed through a service listing, not a statement that the final ERNIE 5.5 release has reached a stable, general-availability status.

The one-million-token context is the headline capability. In practical terms, it gives applications room to pass very large documents, long conversation histories or extensive codebases in one request, although useful performance will still depend on the task, prompt design, output limits and the account’s actual quota. A large context window is an input capacity, not proof that every answer will remain equally reliable across that entire span.

Baidu’s related vLLM-ERNIE project describes ERNIE-5.5-Flash as a sparse mixture-of-experts model with 280 billion parameters and 14 billion activated parameters per token. The repository describes chat, reasoning and tool-calling support as forthcoming, while released Base checkpoints are available. Those repository details should not be confused with the Qianfan Preview’s service guarantees.

The release position

The official Qianfan page confirms the endpoint and its published limits, but it does not give a final launch date for a wider release or confirm the open-weight timetable reported elsewhere. Pandaily has reported that Baidu plans to release open weights on 25 November, but that date is not independently confirmed by the official model-list page. ¹

That distinction is important for developers deciding whether to build around the model. Qianfan access and open-weight availability are separate routes, with different deployment, pricing and operational trade-offs. The current official evidence supports saying that ERNIE 5.5 Flash Preview is listed on Qianfan with a one-million-token context; it does not support treating every roadmap claim as settled.

Our opinion

A one-million-token context window is useful when a model can retrieve the right detail without drowning in irrelevant material, so the number alone is not the story. The more meaningful test will be how ERNIE 5.5 Flash handles long documents, tool calls and sustained reasoning under real Qianfan limits. For now, Baidu’s listing makes the Preview worth watching, while the open-weight date and broader release claims remain separate promises rather than confirmed facts.