NOUMENOpen app
NOUMEN

upstage · Efficient

Solar Mini 4

Added Sep 23, 2026

Compare models

Choose up to 3 models to compare.1 of 3 chosen

Loading models…

Use in the app

Upstage's compact 35B-parameter mixture-of-experts model with 3B active parameters and a 524K context window. Built for fast, cost-efficient agentic tasks, with strong Korean and English support.

  • Tools
  • Structured output
Input
Text
Output
Text
Discovery tier
Efficient

Pricing

USD per 1M tokens
Input
$0.05
Output
$0.20
Cache read
$0.005

Reference prices as published by the model provider. API example

Specifications

Context window
524.3K
Max output
131.1K
Average output (7 days)
561 tokens
Longer than 53% of 468 models

Side by side

Solar Mini 4 compared

Specifications, prices, speed and benchmark scores next to each other. Benchmarks below show every compared model too.

Compare models

Choose up to 3 models to compare.1 of 3 chosen

Loading models…

Solar Mini 4upstageThis modelDeepSeek V4 Flash 0423 (Thinking)DeepSeekRemove
ProviderupstageDeepSeek
Discovery tierEfficientOpen weights
AddedSep 23, 2026Oct 7, 2026
Limits
Context window524.29K1.05MBest
Max output131.07K—
Average output561 tokens1.1K tokens
Pricing, USD per 1M tokens
Input$0.05Best$0.14
Output$0.20Best$0.28
Cache read$0.005Best$0.028
Performance
Throughput90.6 tok/s102.8 tok/sBest
Time to first token2.2 s1.8 sBest
Request success (24h)89.1%96.7%Best
Benchmark indexes
Intelligence Index24.124.4Best
Coding Index—52.0
Capabilities
ReasoningNoYes
ToolsYesYes
Structured outputYesYes
Reasoning effort—none, low, high, max

Public evaluations

Benchmarks

Independent evaluation results, with how each score ranks among the models in the Noumen catalog.

  • Solar Mini 4
  • DeepSeek V4 Flash 0423 (Thinking)

Intelligence Index

  • Solar Mini 424.1
  • DeepSeek V4 Flash 0423 (Thinking)24.4

Solar Mini 4: Better than 59% of 326 scored models

3.8Catalog models by score55.1

Indexes

  • Coding Index

    Composite of the coding evaluations

    • Solar Mini 4: —
    • DeepSeek V4 Flash 0423 (Thinking): 52.0

Agentic work

  • AutomationBench-AA

    Workflow automation, penalized for guardrail violations

    • Solar Mini 4: 22.3%
    • DeepSeek V4 Flash 0423 (Thinking): 28.5%Top

  • AutomationBench-AA tasks completed

    Workflows finished end to end without a guardrail violation

    • Solar Mini 4: 2.4%
    • DeepSeek V4 Flash 0423 (Thinking): 8.7%Top

  • AA-Briefcasev1.1

    Agentic knowledge work, rated head to head

    • Solar Mini 4: 867 Elo
    • DeepSeek V4 Flash 0423 (Thinking): 1,041 EloTop

  • GDPval-AAv2.1

    Economically valuable occupational tasks, rated head to head

    • Solar Mini 4: 1,085 EloTop
    • DeepSeek V4 Flash 0423 (Thinking): 1,000 Elo

Document reasoning

  • GDP.pdf

    Professional reasoning over PDF documents, all-pass rate

    • Solar Mini 4: 5.0%
    • DeepSeek V4 Flash 0423 (Thinking): 10.4%Top

  • AA-LCRv1.1

    Reasoning across long documents

    • Solar Mini 4: 83.3%Top
    • DeepSeek V4 Flash 0423 (Thinking): 72.0%

  • MLCR-AA

    Long-context reasoning over medical records

    • Solar Mini 4: 6.7%
    • DeepSeek V4 Flash 0423 (Thinking): —

Reasoning and knowledge

  • Humanity’s Last Exam

    Expert-level questions across academic fields

    • Solar Mini 4: 25.8%
    • DeepSeek V4 Flash 0423 (Thinking): 30.3%Top

  • GPQA Diamond

    Graduate-level science questions

    • Solar Mini 4: —
    • DeepSeek V4 Flash 0423 (Thinking): 86.7%

  • IFBench

    Following precise output instructions

    • Solar Mini 4: —
    • DeepSeek V4 Flash 0423 (Thinking): 73.5%

Coding

  • Terminal-Benchv4.0

    Practical software engineering in a terminal

    • Solar Mini 4: 1.0%
    • DeepSeek V4 Flash 0423 (Thinking): 3.0%Top

  • SciCode

    Scientific computing in Python

    • Solar Mini 4: 47.6%Top
    • DeepSeek V4 Flash 0423 (Thinking): 40.2%

Legacy evaluations

  • Terminal-Bench Hard

    Earlier terminal task suite

    • Solar Mini 4: —
    • DeepSeek V4 Flash 0423 (Thinking): 38.6%

  • τ²-Bench Telecom

    Tool use in a customer-support simulation

    • Solar Mini 4: —
    • DeepSeek V4 Flash 0423 (Thinking): 95.6%

  • AA-LCR (unversioned)

    Long-context reasoning, earlier grading

    • Solar Mini 4: 83.3%Top
    • DeepSeek V4 Flash 0423 (Thinking): 72.0%

Scores from Artificial Analysis, updated Oct 26, 2026. Artificial Analysis

Live routing data

Performance and reliability

Recent averages across requests routed for this model. Your own speed depends on prompt length, reasoning effort and load.

Throughput
90.6 tok/s
Faster than 59% of 253 models
Time to first token
2.2 s
Quicker than 72% of 253 models
Prompt cache hits
22.9%
About $0.0397 per 1M input tokens with caching
Request success (24h)
89.1%
Unstable

Cache hits are the share of input tokens recently served from a prompt cache across all requests. They are not a forecast of your own hit rate.

Hourly request success · last 72 hoursLowest hour 0.0%
  • Healthy (98% or more)
  • Degraded (90–98%)
  • Down (under 90%)
  • No requests

Developers

Use it through the API

Noumen’s API is OpenAI-compatible: any Chat Completions client works with a base URL and a key.

Read the quickstart
curl https://www.noumen.xyz/api/v1/chat/completions \
  -H "Authorization: Bearer $NOUMEN_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "upstage/solar-mini4",
    "messages": [{ "role": "user", "content": "Explain this stack trace." }]
  }'
import os
from openai import OpenAI

client = OpenAI(base_url="https://www.noumen.xyz/api/v1", api_key=os.environ["NOUMEN_API_KEY"])
reply = client.chat.completions.create(
    model="upstage/solar-mini4",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

upstage

Other upstage models, starting with the same family.