Atlas

Models

Baidu

ERNIE 5.5 Flash Preview

162.9
Capability · 95% interval 161.1–164.9
Rank 23 of 895 (interval 18–34)

ERNIE 5.5 Flash Preview is Baidu's first post-trained model built on ERNIE 5.5 Flash Base, released November 11, 2026 on the Qianfan API as `ernie-5.5-flash-preview`. Baidu describes a text-only sparse Mixture-of-Experts model with 280B total and 14B active parameters (rounded) and a 1M-token context, without an exact limit, and calls it an intermediate post-training checkpoint ahead of the official ERNIE 5.5 Flash release. Qianfan lists it among both its text-generation and its deep-thinking models, which output a chain of thought before the answer, and documents no thinking switch or effort levels. Baidu says the weights will be open-sourced under Apache 2.0; they are not yet published.

Scored on 16 benchmarks · 15 total scores · 0 independently reported

Score providers

Organizations reporting headline scores for 16 benchmarks

In fitin the capability model (8 of 16)
Agents · 4 benchmarks
AutomationBench (Public Set)
2 / 21——2026-11-11
MCP Atlas April 2026 (public 500 tasks)
2 / 10——2026-11-11
Toolathlon-VerifiedIn fit
8 / 42~2 mo~Feb 20272026-11-11
Agents' Last Exam ALE-CLI Pass Rate
8 / 34——2026-11-11
Code · 4 benchmarks
NL2Repo - release-reported repository generation scoreIn fit
5 / 33~2 mo~May 20272026-11-11
ProgramBench - almost solved rollout percentage
8 / 14——2026-11-11
Terminal-Bench 4.0In fit
15 / 69~1 mo~Apr 20272026-11-11
DeepSWE v1.1In fit
21 / 78~3 mo~Apr 20272026-11-11
Search · 3 benchmarks · first place on 1
DRACO - release-reported research score
1 / 2——2026-11-11
BrowseCompIn fit
16 / 79~5 moSep 20262026-11-11
AA-LCR v1.1In fit
70 / 408~6 mo~Mar 20272026-11-11
Professional Work · 2 benchmarks · first place on 2
OfficeQA Pro v2 (release-reported)
1 / 2——2026-11-11
SpreadsheetBench 2In fit
1 / 16~2 mo~Sep 20282026-11-11
General QA · 1 benchmark
Humanity's Last Exam (Text-Only)
27 / 490——2026-11-11
Indexes · 1 benchmark
Capability
162.9 / 171Capability
23 / 895——2026-11-13
Science · 1 benchmark
Terminal-Bench-Science 0.1In fit
17 / 74~1 mo~Mar 20272026-11-11
Loading Atlas data…