Model

Reka Flash 3

OPEN

Reka AI · Reka family · released Mar 10, 2025

21B from-scratch reasoning model with a novel budget-forcing mechanism; text-only despite the Reka name's multimodal history; rivals OpenAI o1-mini.

ReasoningCodingVisionFunction callingTool useAgentic

1241.4

Elo · rank #170

Parameters

21B

Active params

21B (dense)

Context

32K tokens

Architecture

21B dense transformer, reasoning model with 'budget forcing' CoT control, Llama-compatible

License

Apache 2.0

Languages

—

API price (in/out)

$0.2 / $0.8

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

BFCLAgents77.2%#15

best: SmolLM3 3B (No Thinking) · 92.3%

GPQA DiamondReasoning52.9%#181

best: GPT-5.6 · 94.6%

Humanity's Last ExamReasoning5.1%#117

best: Claude Sonnet 5 · 57.4%

HumanEvalCoding96.3%#5

best: Claude Opus 4.5 · 99.4%

IFBenchReasoning30.4%#66

best: MiniMax M3 · 83.0%

LiveCodeBenchCoding62.0%#96

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MATH-500Math90.6%#45

best: GPT-5 · 99.4%

MMLU-ProKnowledge65.0%#109

best: Claude Fable 5 · 91.5%

Run it locally

VRAM @ Q4

14 GB

VRAM @ FP16

42 GB

Fits on (Q4)

RTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Throughput data unavailable.

Quantizations

GGUF Q4

Reka family

Elo progression across releases

Reka Edge (7B)Feb 2024378.0 Reka Flash (21B)Feb 2024633.5 Reka CoreApr 2024909.9 Reka Flash 3Mar 20251241.4 Reka Edge (2026)Mar 20261698.3

Resources

HFRekaAI/reka-flash-3 DOCSdocs.reka.ai APIAvailable via Reka AI API

API price $0.2/$0.8 · each benchmark row carries its own source badge (see methodology)