Baidu's ERNIE 5.5 Flash Preview enters public testing on the LMSYS Arena
11/11/2026 — 11/12, 22:21·5 sources (all)·10 reports (all)
Key takeaways
- Baidu's ERNIE 5.5 Flash Preview is now free to try on OpenRouter and has appeared on the LMSYS Arena for public testing, where community eva…
Background & analysis
On November 11, 2026, Baidu officially released ERNIE 5.5 Flash Preview, a foundation model the company described as its latest open-weight release in the ERNIE family, built on a new architecture designed for efficient inference. Reporting from Leiphone and 36Kr put the model at 280B total parameters with 14B active per token and support for up to 1 million token context. Baidu said it performs well in coding, 3D modeling, knowledge work, and content creation, posted competitive results on benchmarks such as Terminal-Bench and AutomationBench, and was specifically trained for game development with compatibility with mainstream development tools.
Later that day, Baidu announced that ERNIE 5.5 Flash Preview was available for free trial on OpenRouter. Zhidongxi reported that the model is the first post-training version of the ERNIE-5.5-Flash-Base foundation model, generates at over 200 tokens per second, and that its API had already gone live on the Qianfan platform, with weights scheduled for open-sourcing before November 25. The pricing uses time-of-day tiers: peak-hour input at CNY 1 per 1M tokens and output at CNY 4, versus list prices of CNY 2 and CNY 8; idle-hour rates are half. Baidu also claimed the model outperformed DeepSeek-V4.1-Flash-1017 on 14 benchmarks.
On November 12, Baidu invited developers to test the preview with difficult tasks and announced that ERNIE 5.5 Flash Preview had appeared on the LMSYS arena for public testing. Leiphone noted this was the third Wenxin release within a week, after the Wenxin 5.5 Flash Base foundation model was open-sourced on November 5. Zhidongxi confirmed the launch at Baidu World 2026, where the company highlighted coding, knowledge work, and design creation, and released two Wenxin application models. That evening, TMTPost published hands-on testing, calling the model strong in coding, 3D modeling, knowledge work, and content creation. The model remains in preview and public-testing status, with the open-weight release targeted before November 25.
AI-generated from 10 reports · updated 22 hours ago
Latest turnTMTPost tested Baidu's newly released ERNIE 5.5 Flash Preview and reported strong performance in coding, 3D modeling, knowledge work, and content creation, with its aesthetic quality still solid. The hands-on report did not provide benchmark scores or public availability details. The model is a preview release from Baidu.
Related AI tools
- Heat index
- 166
- All sources
- 5
- All reports
- 10
- First seen
- 2 days ago
Reports on this story headlines open the original
TMTPost tested Baidu's newly released ERNIE 5.5 Flash Preview and reported strong performance in coding, 3D modeling, knowledge work, and content creation, with its aesthetic quality still solid. The hands-on report did not provide benchmark scores or public availability details. The model is a preview release from Baidu.
钛媒体AI score 59At its World conference, Baidu released ERNIE 5.5 Flash Preview with 280 billion parameters plus two application-focused ERNIE models, saying the update targets coding, knowledge work and design. The event also brought upgrades to Baidu's consumer apps and new infrastructure, including a 1,024-card Tianchi supernode and an updated Kunlun chip roadmap.
智东西AI score 63Baidu released ERNIE 5.5 Flash Preview, its third ERNIE move in a single week, after open-sourcing the ERNIE 5.5 Flash Base model on Nov 5. According to Leiphone, the push is built around three pitches: faster, cheaper and open source.
雷峰网AI score 55Baidu's ERNIE 5.5 Flash Preview model has appeared on the LMSYS Arena for public evaluation. The model utilizes a MoE architecture featuring 280 billion total parameters and 14 billion active parameters per token.
X：百度 (@Baidu_Inc)First-partyAI score 53Baidu has announced the release of the ERNIE 5.5 Flash Preview, inviting users and developers to test the model with challenging tasks. The new preview delivers enhanced inference performance and is now available for evaluation.
X：百度 (@Baidu_Inc)First-partyAI score 55
Baidu has released ERNIE 5.5 Flash Preview, a base model with 280B total parameters, 14B parameters activated per token, and context of up to 1M tokens, according to 36Kr. It is the company's newest base model and is available as a preview.
36氪AI score 62Baidu has released ERNIE 5.5 Flash Preview, the first post-trained version of its ERNIE-5.5-Flash-Base model, with 280B total parameters, 14B active parameters, 1M-token context and over 200 tokens/sec generation, according to Zhidx. The API is live on Baidu's Qianfan platform and weights are due to be open-sourced before November 25, priced at 1 yuan per 1M input tokens and 4 yuan per 1M output tokens during peak hours. Baidu says the model beats DeepSeek-V4.1-Flash-1017 on 14 benchmarks.
智东西AI score 54Baidu announced on X that ERNIE 5.5 Flash Preview is now available to try for free on OpenRouter. The company shared no parameters, context length, or general-availability date for the preview. Developers can test it directly through OpenRouter.
X：百度 (@Baidu_Inc)First-partyAI score 50Baidu announced ERNIE 5.5 Flash Preview, the latest open-weight foundation model in the ERNIE family, built on a new architecture aimed at efficient inference. The company says it delivers stronger performance in coding, 3D modeling, knowledge work and content creation while keeping decoding fast.
X：百度 (@Baidu_Inc)First-partyAI score 59Baidu has released ERNIE 5.5 Flash Preview, a base model with 2800B total parameters and 140B activated per token, supporting up to a 1M token context window. Baidu says it performs strongly on coding, 3D modeling, knowledge work and content creation, posts competitive scores on benchmarks including Terminal-Bench and AutomationBench, and was specifically trained and adapted for game development workflows.
雷峰网AI score 59
Other stories people are talking about
- Heat index 314RisingVolcano Engine launches Seedance 2.5 flash with lower pricing5 sources
- Heat index 95NewDeveloper Introduces LLM Gateway Implementation Using hikyaku1 sources
- Heat index 78DeepSeek Harness Launches iPhone Client DeepPilot in TestFlight1 sources
- Heat index 83Space AI pioneer breaks on debut after reporting three-year losses of 570M CNY1 sources
- Heat index 70OpenRouter Publishes Tutorial on Using Rerankers to Fix RAG Retrieval Errors1 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion