ERNIE-5.5-Flash (280B MoE) Architecture Parameters, Benchmarks & VRAM
Baidu ERNIE 5.5 sparse MoE (286B total / 14B active, 512 experts) supporting 1M context.
🧬 ERNIE-5.5-Flash (280B MoE) Architecture & Tensor Specifications
Verified directly against HuggingFace (baidu/ERNIE-5.5-Flash-Base) config.json & Model Card
1. Backbone & Dimensions
ernie5_5_moe- Total / Active Params
- 286B / 14B
- Hidden Size (d_model)
- 4,096
- FFN Intermediate Size
- 14,336
- MoE Expert Dim
- 2,048
- MoE Experts
- 512 Routed · Top-10 + 1 Shared
- Activation / Norm
- SiLU (SwiGLU) · ε=0.000001
- Native Tensor Dtype
- BF16 / FP8
2. Attention & Layer Layout
MLA- Num Hidden Layers
- 43 Layers
- Hidden Layer Layout
- 43-Layer Sparse MoE (512 Experts, Top-10 Active)
- Attention Heads (Q / KV)
- 64 Q / 1 KV (64:1 GQA)
- Head Dimension
- 512
3. Context, RoPE & Tokenizer
Tokenizer- Native Context Length
- 1,048,576 tokens
- RoPE Configuration
- RoPE (原生 1,048,576 Context)
- Vocab Size
- 201,216 (Untied)
📋 ERNIE-5.5-Flash (280B MoE) Official API & System Card Specifications
Verified against official Baidu 百度文心 developer documentation & pricing
1. Identity & Capabilities
API Spec- API Model ID
- ernie-5.5-flash-preview
- Release Date
- 2026-11-10
- Weight License
- Open-Weight (API Spec)
- Native Reasoning
- ✅ Supported
- Function Calling & JSON Schema
- ✅ 原生支持
- Batch API 异步批处理
- ✅ Supported (50% off)
2. Context, Output & Modalities
Limits- Max Context Window
- 1,024,000 tokens (1024K)
- Max Output Tokens
- 64,000 tokens
- Input Modalities
- Text
- Output Modalities
- Text
- Direct China Access
- ✅ Direct CN + RMB
- Deployment Mode
- Open-Weight + Cloud API
3. Official Pricing & Cache
CNY / 1M- Input / 1M Tokens
- ¥ 0
- Cached Input / 1M
- ¥ 0
- Output / 1M Tokens
- ¥ 0
- Tiered Pricing Notes
- Baidu Qianfan: Free preview tier for Baidu ERNIE 5.5 Flash (1M ctx)
🎛️ Official Sampling & Reasoning Parameters
Recommended generation settings & API parameter constraints
📊 ERNIE-5.5-Flash (280B MoE) Full Benchmark Scorecard
13 public benchmarks · ranks among 154 models · synced 2026-11-20
💻 Coding 5 items
MCP Atlas is a benchmark for evaluating AI models on scaled tool use capabilities, measuring how well models can coordinate and utilize multiple tools across complex multi-step tasks.
DeepSWE 1.1 evaluates software engineering agents using the mini-swe-agent harness.
NL2Repo evaluates long-horizon coding capabilities including repository-level understanding, where models must generate or modify code across entire repositories from natural language specifications.
Terminal-Bench 4.0 evaluates AI agents on real-world work in terminal and command-line containerized environments across 66 tasks, with emphasis on science-adjacent and frontier engineering problems.
Terminal-Bench-Science 0.1 is a Stanford-led community benchmark of 70 tasks drawn from scientific research workflows across the life, physical, earth, mathematical, and engineering sciences.
🤖 Agents & Tools 7 items
BrowseComp is a benchmark comprising 1,266 questions that challenge AI agents to persistently navigate the internet in search of hard-to-find, entangled information.
DRACO is a deep research benchmark that evaluates an agent's ability to gather, synthesize, and reason over information to answer complex research questions. Scores are based on official rubrics per question, with the final score being the average across all questions.
OfficeQA Pro evaluates AI models on professional knowledge-work questions and tasks drawn from real office workflows, including document analysis, spreadsheet reasoning, and information synthesis across business domains.
APEX-Agents is a benchmark evaluating AI agents on long horizon professional tasks that require sustained reasoning, planning, and execution across complex multi-step workflows.
AutomationBench is a tool-use benchmark that evaluates AI agents on automating real-world workflows, testing their ability to orchestrate tools and complete multi-step automation tasks.
SpreadsheetBench 2 evaluates office automation agents on spreadsheet analysis, reasoning, and manipulation tasks.
Agents' Last Exam is a challenging benchmark for AI agents on hard, long-horizon tasks that test sustained reasoning, planning, and tool use, reported with and without tool access.
Source: ZeroEval / llm-stats.com (✓ = third-party verified, others vendor-reported).
⚡ ERNIE-5.5-Flash (280B MoE) Interactive VRAM & Token Speed Calculator
Physical 512 GB · Load 38.7%
MoE: reads only 8.49 GB active weights/step
MLA compresses multi-batch KV footprint
OLLAMA_CONTEXT_LENGTH=16384 ollama run baidu-ernie-5-5-flashMac Studio M5 Ultra 512G Fits ERNIE-5.5-Flash (280B MoE) · Cost & TCO
Approx. $9,499 (~$19/GB VRAM) · 190W TDP
📊 ERNIE-5.5-Flash (280B MoE) Quantization VRAM Breakdown (16K Context)
| Quantization | Bits/Weight | Weights | KV Cache (16K) | Total VRAM | Quality Note |
|---|---|---|---|---|---|
| FP16 / BF16 | 16 bpw | 572 GB | 0.81 GB | 584.7 GB | Lossless Full Precision |
| FP8 / Q8_0 | 8.5 bpw | 303.88 GB | 0.81 GB | 311.21 GB | 99.9% Near-Lossless |
| GGUF Q6_K | 6.56 bpw | 234.52 GB | 0.81 GB | 240.47 GB | Ultra High Quality |
| GGUF Q5_K_M | 5.69 bpw | 203.42 GB | 0.81 GB | 208.75 GB | High Balanced Quality |
| Q4_K_M / AWQ | 4.85 bpw | 173.39 GB | 0.81 GB | 178.12 GB | Sweet Spot (Recommended) |
| GGUF IQ4_XS | 4.25 bpw | 151.94 GB | 0.81 GB | 156.24 GB | Compact 4-bit |
| GGUF Q3_K_M | 3.91 bpw | 139.78 GB | 0.81 GB | 143.84 GB | Moderate Perplexity Loss |
| GGUF Q2_K | 3.15 bpw | 112.61 GB | 0.81 GB | 116.13 GB | Extreme Compression |
🖥️ GPU & Apple Silicon Compatibility for ERNIE-5.5-Flash (280B MoE) (Q4_K_M)
Frequently Asked Questions
What are the exact architecture parameters of ERNIE-5.5-Flash (280B MoE)?
ERNIE-5.5-Flash (280B MoE) uses Ernie5_5_MoeForCausalLM with 43 layers (43-Layer Sparse MoE (512 Experts, Top-10 Active)), hidden_size=4096, 64 Q heads / 1 KV heads (head_dim=512), and vocab_size=201,216.
How much VRAM does ERNIE-5.5-Flash (280B MoE) require?
At 4-bit quantization (Q4_K_M, 16K context), ERNIE-5.5-Flash (280B MoE) requires 178.12 GB VRAM. Full FP16 requires 584.7 GB.