Tasks and inputs
ERNIE 5.5 Flash Preview is listed for Language & text, Text → Text, Coding & Code Generation. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 17, 2026
ERNIE 5.5 Flash Preview is Baidu's first post-trained 280B-A14B text model, with a 1M window, aimed at game dev, 3D scripting and knowledge work.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: No price published; a free variant exists on at least one third-party aggregator. Confirm on your own Qianfan account.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Game studios and technical artists are the unusual first audience Baidu picked for this one. Its target list starts with Unreal, Unity and Godot work, moves through Blender Python and FreeCAD scripting, and only then reaches spreadsheets, slide decks and research reports.
It is an intermediate checkpoint. Baidu released it on 11 November 2026 as the first post-trained model built on ERNIE 5.5 Flash Base, says post-training is still under way, and expects it to fall short in some areas until the final release.
| Field | Value |
|---|---|
| Architecture | Sparse MoE, 280B total / 14B active |
| Context | 1M tokens |
| Modality | Text in, text out |
| Hosted access | Baidu Qianfan (model page am-abe18d298d70); free route on a major aggregator |
| Price | Not published |
| Weights | Promised under Apache 2.0; not yet released |
Demos of audio and video output were produced by the model writing code and calling tools, not by native multimodal generation. Plan on text only.
Is there a hosted API? Baidu's release page links to Qianfan; at least one early write-up said no hosted API accompanied the Preview. The Qianfan link is the first-party statement, so check your Qianfan console and confirm the exact model identifier there, since the page ID is not necessarily the API name.
Which licence? The Preview's weights are promised under Apache 2.0, while the Base checkpoint uses a custom Baidu licence with regional restrictions. Do not assume the two match.
When do the weights land? Baidu said "soon"; one report gave 25 November 2026. Until a repository appears, this is hosted-only.
Pick a task from its target list that you already do by hand. A good one: ask it to write a Blender Python script that builds a parametric object, run the script, and fix errors in a loop. Or give it a small Godot project and a feature request. Count how many rounds it needs to reach working code. Repeat the same tasks with your current coding model so you can compare.
Because this is a preview, rerun the suite after the final release; results may shift.
For long-form coding agents with open weights already available, Meituan's LongCat line (see LongCat-2.5-Preview) is one alternative. Baidu's open checkpoints are on the Baidu Hugging Face page, and the code-generation directory lists more.
Compare the model on the work you need to complete.
ERNIE 5.5 Flash Preview is listed for Language & text, Text → Text, Coding & Code Generation. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Documented limits: 1M context. Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Meta
Muse Ember is Meta's smallest, cheapest Muse model: a 1M-context multimodal worker for tool calls, coding subtasks, routing and extraction at volume.
iflytek
Spark X2.6 is iFLYTEK's 293B mixture-of-experts language model family with integrated device and cloud variants, unveiled in October 2026.
Meta
Muse Spark Open is Meta's open-weight 816B mixture-of-experts model (42B active) with 1M context, built for coding and agents you can self-host.
ByteDance Seed
Seed 2.1 Lite is a lightweight Doubao Seed 2.1 model listed on Volcengine Ark with a 1M context, 256K output, audio and image input and tool calls.