[whatmodel.sh](../../README.md) / Reka Flash 21B (2024-02-26) # Reka Flash 21B (2024-02-26) Reka AI · released 2024-02 · closed weights **score 44.4** ±0.8 · comparison only: no current run option verified · from 11 benchmark results across 6 sources Score within the [LMArena (text)](https://whatmodel.sh/?set=lmarena-text&score=set) comparison set only. ## What it's good at ● marks Reka Flash 21B (2024-02-26) among every model on that task or test, worst to best. 
task                vs the others          rank  counts  tests
coding              ▁▁▁▃▄●█▇▅▄▃▁   #423 of 539*     17%      1
chat & writing      ▁▁▁▃▅●▇█▇▇▇▃   #352 of 474*     12%      2
math                ▁▁▃●██▅█▃▁▁▁   #469 of 530*     12%      1
vision & documents  ▁▁●▃▂▃▅▄▆██▄    #181 of 189     17%      4
coding · #423 of 539 · 17% of its score Beats [LFM2.5 8B-A1B](lfm2-5-8b-a1b.md), [Qwen3-VL 2B Thinking](qwen3-vl-2b-thinking.md), [DeepSeek R1 Distill Qwen 7B](deepseek-r1-distill-qwen-7b.md) here, though they rank higher overall. Trails [Code Tulu 2 13B](code-tulu-2-13b.md), [Phi-3-mini-128k-instruct (Apr 2024)](phi-3-mini-128k-instruct.md), [DeepSeek LLM 67B Chat](deepseek-llm-67b-chat.md) here, though they rank lower overall. 
test                               result  vs the others in its run  difficulty        counts  why
CRUXEval (ZeroEval, 800 problems)   34.1%  ▁●▆▃▆█▆▃▂·▁▁ beats 2/47   medium · avg 57%     17%  sharp · at its level
chat & writing · #352 of 474 · 12% of its score Beats [Ministral 3 14B Reasoning 2512](ministral-3-14b-reasoning.md), [Gemma 4 E4B](gemma-4-e4b.md), [OpenThinker3 7B](openthinker3-7b.md) here, though they rank higher overall. Trails [EXAONE 3.5 2.4B Instruct](exaone-3-5-2-4b.md), [LFM2 700M](lfm2-700m.md), [Flan-T5 XXL](flan-t5-xxl.md) here, though they rank lower overall. 
test                                result  vs the others in its run   difficulty  counts  why
LMArena (text)                      1226.9  ▁▁▁▂▃●▃▆▆▇█▃ beats 81/385  avg 1260      6.3%  blunt
Polyrating rating (bias-aware re-…  1170.0  ▂▁▂▁▃▇█▆●▄▂▂ beats 74/95   avg 1187      5.8%  typical
math · #469 of 530 · 12% of its score Beats [Yi 34B Chat](yi-34b-chat.md) here, though they rank higher overall. Trails [GPT-Neo 125M](gpt-neo-125m.md), [GPT-Neo 1.3B](gpt-neo-1-3b.md), [OPT-1.3B](opt-1-3b.md) here, though they rank lower overall. 
test                         result  vs the others in its run  difficulty           counts  why
GSM8K (ZeroEval, zero-shot)   74.7%  ▁▄·▁●█▅▅▅███ beats 5/43   very easy · avg 87%     12%  sharp
vision & documents · #181 of 189 · 17% of its score Beats [Gemma 3 4B](gemma-3-4b.md) here, though they rank higher overall. Trails [Qwen3.5 0.8B](qwen3-5-0-8b.md), [North Micro Vision Instruct (2.4B)](north-micro-vision-instruct.md), [Molmo 7B-D (0924)](molmo-7b-d.md) here, though they rank lower overall. 
test                                result  vs the others in its run  difficulty      counts  why
CharXiv descriptive questions        56.5%  ▄●▄▄▄·██·▄·█ beats 1/12   easy · avg 69%    8.3%  at its level
CharXiv descriptive questions (te…   54.7%  ●··▄··█··▄·▄ beats 1/6    easy · avg 69%    4.1%  at its level
CharXiv (reasoning questions)        26.6%  ▂●▂█▃▂·▂▂··▂ beats 1/13   hard · avg 34%    3.2%  typical
CharXiv reasoning questions (test…   27.1%  ▄··●▄▄█····▄ beats 1/6    hard · avg 30%    1.7%  typical
other tests (3) 
test                                          result  vs the others in its run  difficulty           counts  why
MMLU-Redux (ZeroEval, 2,778 quest…             64.7%  ▁·▁▁●▄▄▅█▄▂▁ beats 7/53   easy · avg 75%          31%  at its level
WildBench v2 WB-Score (task macro)  30.3636 WB-Score  ▁·▁▁·▃●▄▇▆▅▆ beats 16/52  avg 43.019 WB-Score    6.7%  sharp
ZebraLogic (ZeroEval, 1,000 puzzl…              9.3%  ▁▁●▆██▁▁·▁▁▁ beats 8/57   hard · avg 15%         4.1%  sharp · too hard for it
## How the score was worked out Each result implies a score on its own, after adjusting for how hard that test is and who ran it. The model's score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its *counts* column above; here it is added up by who ran the tests. [Method](../method.md) 
source                                    results  share of the score
WildEval ZeroEval published results             4   64% ▇▇▇▇▇▇▇▇
CharXiv leaderboard                             2   11% ▇
WildBench v2 leaderboard data                   1    7% ▇
LMArena leaderboard dataset                     1    6% ▇
CharXiv: Charting Gaps in Realistic Cha…        2    6% ▇
Polyrating: A Cost-Effective and Bias-A…        1    6% ▇
prior: pull toward the middle                        0% ▇
weighted average                                   = 44.4
Licences: WildEval ZeroEval published results ([Apache-2.0](../../licenses/zeroeval-published-results-Apache-2.0.txt)) · CharXiv leaderboard ([Apache-2.0](https://www.apache.org/licenses/LICENSE-2.0)) · WildBench v2 leaderboard data ([Apache-2.0](https://github.com/allenai/WildBench/blob/d6b8dcaf377d173d031980f97c16e1a82618c03d/LICENSE)) · LMArena leaderboard dataset ([CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)) · CharXiv: Charting Gaps in Realistic Chart Understanding in Multimodal LLMs ([CC BY-SA 4.0](https://creativecommons.org/licenses/by-sa/4.0/)) · Polyrating: A Cost-Effective and Bias-Aware Rating System for LLM Evaluation ([CC-BY-4.0](https://creativecommons.org/licenses/by/4.0/)) ## How to run it No verified API or local download yet. [interactive version](https://whatmodel.sh/model/reka-flash-21b-20240226) · [JSON](https://whatmodel.sh/api?format=json&all=1)