Skip to main content
Back to Explore

Reka Flash 3

Rekaai·Released 2025-03-12
66K ctx

About

Reka Flash 3 is a general-purpose, instruction-tuned large language model with 21 billion parameters, developed by Reka. It excels at general chat, coding tasks, instruction-following, and function calling. Featuring a...

Related Models

Pricing

Input

$0.10

per 1M tokens

Output

$0.20

per 1M tokens

Blended

$0.13

per 1M tokens

Cheaper than 80% of models. Median price is $0.54/1M tokens.

Cost Calculator

Tokens per day1M
100K100M

Daily

$0.13

Monthly

$3.75

vs. Similar Models

Olmo 3 7B ThinkQ:-0.1
$0.00-100%
Llama 3.2 Instruct 3BQ:+0.1
$0.15+20%
Anthropic: Claude 3 HaikuQ:-0.2
$0.50+300%
Llama 2 Chat 7BQ:+0.2
$0.10-20%

Performance

95

tokens/sec

Faster than 51% of models

1.73

seconds

Faster than 27% of models

22.79

seconds

Faster than 23% of models

Market Median

94 tok/s

1% faster

Median TTFT

1.11s

56% slower

Throughput/Dollar

760

tok/s per $/1M

Speed Comparison

MiMo-V2-Flash (Non-reasoning)
95 tok/s-0%
Qwen3 32B (Non-reasoning)
95 tok/s+0%
Cogito v2.1 (Reasoning)
95 tok/s+0%

Context Window

66K

tokens

Larger than 13% of models

Max Output

66K

tokens

100% of context

Benchmarks

MMLU-Pro
66.9%
GPQA Diamond
52.9%
HLE
5.1%
LiveCodeBench
43.5%
SciCode
26.7%
TerminalBench Hard
0.0%
MATH-500
89.3%
AIME
51.0%
AIME 2025
33.7%
IFBench
30.4%
Long Context Recall
0.0%
Tau2
0.0%
Market AverageTop Score

Quick Compare

Similar Models

Compare all 7 models