NOUMENOpen app
NOUMEN

reka · Open weights

Reka Flash 3

Added Oct 15, 2026Open weights

Compare models

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

Loading models…

Use in the app

Reka's 21B reasoning model for chat and coding, served directly by Reka.

  • Reasoning
  • Structured output
Input
Text
Output
Text
Discovery tier
Open weights

Pricing

USD per 1M tokens
Input
$0.11
Output
$0.21

Reference prices as published by the model provider. API example

Specifications

Context window
65.5K
Max output
65.5K
Average output (7 days)
2.7K tokens
Longer than 95% of 465 models

Side by side

Reka Flash 3 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…

Reka Flash 3rekaThis modelLlama 3.2 3b InstructMetaRemove
ProviderrekaMeta
Discovery tierOpen weightsOpen weights
AddedOct 15, 2026Jan 1, 2024
Limits
Context window65.54K131.07KBest
Max output65.54KBest8.19K
Average output2.7K tokens119 tokens
Pricing, USD per 1M tokens
Input$0.11$0.0306Best
Output$0.21$0.0493Best
Performance
Throughput—160.4 tok/s
Time to first token—0.8 s
Request success (24h)—100.0%
Benchmark indexes
Intelligence Index5.65.7Best
Math Index33.7Best3.3
Capabilities
ReasoningYesNo
Structured outputYesNo
PDF inputNoYes
Reasoning effort——

Public evaluations

Benchmarks

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

  • Reka Flash 3
  • Llama 3.2 3b Instruct

Intelligence Index

  • Reka Flash 35.6
  • Llama 3.2 3b Instruct5.7

Reka Flash 3: Better than 2% of 326 scored models

3.8Catalog models by score55.1

Indexes

  • Math Index

    Composite of the math evaluations

    • Reka Flash 3: 33.7Top
    • Llama 3.2 3b Instruct: 3.3

Document reasoning

  • AA-LCRv1.1

    Reasoning across long documents

    • Reka Flash 3: 0.0%
    • Llama 3.2 3b Instruct: 4.3%Top

Reasoning and knowledge

  • Humanity’s Last Exam

    Expert-level questions across academic fields

    • Reka Flash 3: 4.4%
    • Llama 3.2 3b Instruct: 5.3%Top

  • GPQA Diamond

    Graduate-level science questions

    • Reka Flash 3: 52.9%Top
    • Llama 3.2 3b Instruct: 25.5%

  • MMLU-Pro

    Multi-subject knowledge and reasoning

    • Reka Flash 3: 66.9%Top
    • Llama 3.2 3b Instruct: 34.7%

  • IFBench

    Following precise output instructions

    • Reka Flash 3: 30.4%Top
    • Llama 3.2 3b Instruct: 26.2%

Coding

  • LiveCodeBench

    Recent competitive programming problems

    • Reka Flash 3: 43.5%Top
    • Llama 3.2 3b Instruct: 8.3%

Math

  • AIME 2025

    Competition mathematics, 2025 problems

    • Reka Flash 3: 33.7%Top
    • Llama 3.2 3b Instruct: 3.3%

  • AIME

    Competition mathematics

    • Reka Flash 3: 51.0%Top
    • Llama 3.2 3b Instruct: 6.7%

  • MATH-500

    Competition math problems

    • Reka Flash 3: 89.3%Top
    • Llama 3.2 3b Instruct: 48.9%

Legacy evaluations

  • Terminal-Bench Hard

    Earlier terminal task suite

    • Reka Flash 3: 0.0%
    • Llama 3.2 3b Instruct: —

  • τ²-Bench Telecom

    Tool use in a customer-support simulation

    • Reka Flash 3: 0.0%
    • Llama 3.2 3b Instruct: 21.1%Top

  • AA-LCR (unversioned)

    Long-context reasoning, earlier grading

    • Reka Flash 3: 0.0%
    • Llama 3.2 3b Instruct: 4.3%Top

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

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": "reka/reka-flash-3",
    "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="reka/reka-flash-3",
    messages=[{"role": "user", "content": "Explain this stack trace."}],
)
print(reply.choices[0].message.content)

reka

Other reka models, starting with the same family.