LIVE
BODYBUILDER$-1000000.00▼ 5.4%
MINIMAX-M2.7$1.20▼ 7.2%
QWEN-PLUS-2025$0.78▼ 28.3%
DEEPSEEK-AI$0.00▼ 27.0%
GPT-4.1-NANO$0.40▼ 13.0%
GLM-4-LONG$0.00▼ 14.1%
GROK-4.1-FAST$0.50▼ 16.4%
GROK-4-FAST-RE$0.50▼ 17.1%
GOOGLE$0.00▼ 13.0%
LLAMA-4-MAVERI$0.60▼ 11.9%
GROK-4-FAST-NO$0.50▼ 14.3%
GROK-4-FAST$0.50▼ 6.0%
GROK-4-1-FAST-$0.50▼ 31.2%
MINIMAX-M2.5$0.99▲ 10.7%
GOOGLE$1.50▼ 3.4%
GOOGLE$0.00▼ 27.8%
XAI$0.00▲ 4.5%
-CF$0.00▼ 9.7%
GEMINI-3.1-FLA$1.50▲ 9.4%
MINIMAX-M2.7$0.00▼ 9.9%
GPT-4.1-MINI-2$1.60▼ 31.8%
DEEPSEEK$0.28▼ 2.2%
LYRIA-3-PRO-PR$0.00▼ 29.4%
LYRIA-3-CLIP-P$0.00▼ 22.9%
MINIMAX-M2.1$0.95▲ 2.6%
MINIMAX-01$1.10▲ 8.0%
MINIMAX-M3$1.20▼ 6.1%
GEMINI-2.5-FLA$0.40▼ 28.8%
BODYBUILDER$-1000000.00▼ 5.4%
MINIMAX-M2.7$1.20▼ 7.2%
QWEN-PLUS-2025$0.78▼ 28.3%
DEEPSEEK-AI$0.00▼ 27.0%
GPT-4.1-NANO$0.40▼ 13.0%
GLM-4-LONG$0.00▼ 14.1%
GROK-4.1-FAST$0.50▼ 16.4%
GROK-4-FAST-RE$0.50▼ 17.1%
GOOGLE$0.00▼ 13.0%
LLAMA-4-MAVERI$0.60▼ 11.9%
GROK-4-FAST-NO$0.50▼ 14.3%
GROK-4-FAST$0.50▼ 6.0%
GROK-4-1-FAST-$0.50▼ 31.2%
MINIMAX-M2.5$0.99▲ 10.7%
GOOGLE$1.50▼ 3.4%
GOOGLE$0.00▼ 27.8%
XAI$0.00▲ 4.5%
-CF$0.00▼ 9.7%
GEMINI-3.1-FLA$1.50▲ 9.4%
MINIMAX-M2.7$0.00▼ 9.9%
GPT-4.1-MINI-2$1.60▼ 31.8%
DEEPSEEK$0.28▼ 2.2%
LYRIA-3-PRO-PR$0.00▼ 29.4%
LYRIA-3-CLIP-P$0.00▼ 22.9%
MINIMAX-M2.1$0.95▲ 2.6%
MINIMAX-01$1.10▲ 8.0%
MINIMAX-M3$1.20▼ 6.1%
GEMINI-2.5-FLA$0.40▼ 28.8%
CNReasoningERNIE-4.5-21B- Live · updated daily

ernie-4.5-21b-a3b-thinking

Baidu Wenxin
Efficiency score
89/100
−17.5%
Input price
$0.07/ 1M tokens
Output price
$0.28/ 1M tokens
Context window
131K
Throughput
149tok/s
Availability
96.3%
Cost / task
$0.000

Capabilities

Accepts (input)
Text
Produces (output)
Text

7-day heat trend

−17.5%

Pricing breakdown

Input price
$0.07/ 1M tokens
$0.0001 / 1K
Output price
$0.28/ 1M tokens
$0.0003 / 1K
Blended price
$0.23/ 1M tokens
$0.0002 / 1K

Typical 3:1 output-to-input mix, per 1M tokens

Estimated monthly cost by workload

Chat assistant
$54.60/ mo
1K in · 400 out · 10K req/day
RAG / long context
$109.20/ mo
8K in · 600 out · 5K req/day
Agent / tool use
$151.20/ mo
3K in · 1.5K out · 8K req/day
Estimate your cost →

Market position

  • Cheaper than 38% of tracked models
  • Faster than 53% of tracked models
  • Efficiency rank: #248 of 1105

Best suited for

Complex reasoning, analysis, planning and multi-step problem solving where answer quality matters more than raw cost.

About ernie-4.5-21b-a3b-thinking

ERNIE-4.5-21B-A3B-Thinking is Baidu's upgraded lightweight MoE model, refined to boost reasoning depth and quality for top-tier performance in logical puzzles, math, science, coding, text generation, and expert-level...

ernie-4.5-21b-a3b-thinking is a Reasoning model from Baidu Wenxin (CN). HotON.ai tracks it at $0.07 per 1M input tokens and $0.28 per 1M output tokens, with a 131K-token context window, ~149 tokens/sec throughput and 96.3% availability. Its composite efficiency score is 89/100 at an estimated $0.000 per successful task.

Compare ernie-4.5-21b-a3b-thinking

ERNIE-4.5-21B- vs DEEPSEEK-AIERNIE-4.5-21B- vs -CFERNIE-4.5-21B- vs DOLPHIN3.0-R1-ERNIE-4.5-21B- vs DEEPSEEK-AI

Related market news

Five labs, five minds: building a multi-model finance drama on small models4 hours agoWhat to expect from WWDC 2026: Siri’s highly anticipated revamp and Apple Intelligence updates5 hours agoSriram Krishnan is leaving his role as White House AI advisor6 hours ago

Frequently asked questions

How much does ernie-4.5-21b-a3b-thinking cost per 1M tokens?+

ernie-4.5-21b-a3b-thinking is tracked at $0.07 per 1M input tokens and $0.28 per 1M output tokens. A typical 3:1 output-to-input workload blends to roughly $0.23 per 1M tokens. Figures are illustrative demo data.

What is ernie-4.5-21b-a3b-thinking best for?+

Complex reasoning, analysis, planning and multi-step problem solving where answer quality matters more than raw cost.

How fast is ernie-4.5-21b-a3b-thinking?+

ernie-4.5-21b-a3b-thinking delivers about 149 tokens/sec with 96.3% tracked availability, suitable for latency-sensitive, real-time applications.

Is ernie-4.5-21b-a3b-thinking cheaper than other AI models?+

Within the HotON.ai tracked set, ernie-4.5-21b-a3b-thinking is cheaper than 38% of models on input price and ranks #248 of 1105 by overall efficiency.

Related models

deepseek-ai/DeepSeek-R1
Baseten Model APIs
−6.3%
@cf/deepseek-ai/deepseek-r1-distill-qwen-32b
Cloudflare Workers AI
−29.6%
Dolphin3.0-R1-Mistral-24B
Cognitive Computations
−10.0%
deepseek-ai/DeepSeek-R1
DeepInfra
+2.3%

Pricing is real (via the TestKey catalog, updated daily). Quality (Arena Elo) is real where the model is ranked on LMArena. Speed, availability and efficiency are modeled estimates.