whatmodel.sh / Reka Flash 21B (2024-02-26)
Reka Flash 21B (2024-02-26)
↗ share
+ contribute
Reka Flash 21B (2024-02-26) is a closed-weights language model from Reka AI, released February 2024.
Catalogue identity notes
Comparison-only. Arena entries 'reka-flash-21b-20240226' (text), 'reka-flash-21b-20240226-online' (text). LMArena's leaderboard table (arena-leaderboard space @6f2eeeb) labels 'reka-flash-21b-20240226' as 'Reka-Flash-21B' (Reka AI, Proprietary), linking https://www.reka.ai/news/reka-flash-efficient-and-capable-multimodal-language-models. LMArena's leaderboard table (arena-leaderboard space @6f2eeeb) labels 'reka-flash-21b-20240226-online' as 'Reka-Flash-21B-online' (Reka AI, Proprietary), linking https://docs.reka.ai/http-api.html#generation. Released 2024-02: date in Reka's API snapshot id. LMArena labels it 'Reka-Flash-21B' (link: Reka Flash announcement). No longer among Reka's public models.
$ how it compares
The LLMs ranked either side of Reka Flash 21B (2024-02-26), out of 828 with a score, then more from Reka AI.
| # | model | score | where | ||
|---|---|---|---|---|---|
| 570 |  Code Llama 70B (base) | 43±1 | local | ||
| 571 |  Gemma 3n E2B IT | 43±2 | local | ||
| 572 | Reka Flash 21B (2024-02-26) | 43±1 | comparison only | ||
| 573 |  T5 Small | 43±9 | local | ||
| 574 |  MiniCPM-V 2.6 | 43±2 | local | ||
| more from Reka AI | |||||
| 453 | Reka Core (2024-05-01) | 48±1 | comparison only | ||
$ what it's good at
coding▁▁▁▃▅▇█▇▅▄▃▁ #411 of 524 * · 17% of its score · 1 test
Beats LFM2.5 8B-A1B, Qwen3-VL 2B Thinking, DeepSeek R1 Distill Qwen 7B here, though they rank higher overall. Trails Phi-3-mini-128k-instruct (Apr 2024), Code Tulu 2 13B, DeepSeek LLM 67B Chat here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CRUXEval (ZeroEval, 800 problems) | 34.1% ✎ | ▁▃▆▃▆█▆▃▂·▁▁ beats 2/47 | 17% |
chat & writing▁▁▁▃▅▄▆█▇▇▇▃ #343 of 463 * · 12% of its score · 2 tests
Beats Ministral 3 14B Reasoning 2512, Gemma 4 E4B, OpenThinker3 7B here, though they rank higher overall. Trails Flan-T5 XXL, LFM2 700M, EXAONE 3.5 2.4B Instruct here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Polyrating rating (bias-aware re-fit of Chatbot Arena votes, mid-2024) | 1170.0 ✎ | ▂▁▂▁▃▇█▆▇▄▂▂ beats 74/95 | 5.9% |
| LMArena (text) | 1226.7 ✎ | ▁▁▁▂▃▃▃▆▆▇█▃ beats 81/379 | 5.9% |
math▁▂▃▅▇▅▅█▄▁▁▁ #416 of 462 * · 13% of its score · 1 test
Beats Yi 34B Chat here, though they rank higher overall. Trails Gemma 3 1B, Qwen3 0.6B, Tulu 2 DPO 7B here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| GSM8K (ZeroEval, zero-shot) | 74.7% ✎ | ▁▄·▁▄█▅▅▅███ beats 5/43 | 13% |
vision & documents▁▁▃▃▂▃▅▄▇▇█▃ #174 of 182 · 17% of its score · 4 tests
Beats Gemma 3 4B, CogVLM2 Llama3 Chat 19B here, though they rank higher overall. Trails Qwen3.5 0.8B, Molmo 7B-D (0924), North Micro Vision Instruct (2.4B) here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| CharXiv descriptive questions | 56.5% ✎ | ▄█▄▄▄·██·▄·█ beats 1/12 | 8.0% |
| CharXiv descriptive questions (test split) | 54.7% ✎ | █··▄··█··▄·▄ beats 1/6 | 4.5% |
| CharXiv (reasoning questions) | 26.6% ✎ | ▂▃▂█▃▂·▂▂··▂ beats 1/13 | 3.1% |
| CharXiv reasoning questions (test split) | 27.1% ✎ | ▄··▄▄▄█····▄ beats 1/6 | 1.7% |
other tests3 not in a task · 41% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| MMLU-Redux (ZeroEval, 2,778 questions) | 64.7% ✎ | ▁·▁▁▄▄▄▅█▄▂▁ beats 7/53 | 31% |
| WildBench v2 WB-Score (task macro) | 30.3636 WB-Score ✎ | ▁·▁▁·▃█▄▇▆▅▆ beats 16/52 | 6.7% |
| ZebraLogic (ZeroEval, 1,000 puzzles) | 9.3% ✎ | ▁▁▃▆██▁▁·▁▁▁ beats 8/57 | 4.1% |
$ where to run it
No verified API or download recorded yet · add one
$ how the score was worked out · WildEval ZeroEval published r… 64%, CharXiv leaderboard (princeto… 11%, WildBench v2 leaderboard data… 7% and 3 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The 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's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 WildEval ZeroEval published results | 4 | 64% |
 CharXiv leaderboard (princeton-nlp/CharXiv README) | 2 | 11% |
| WildBench v2 leaderboard data (Ai2) | 1 | 7% |
 CharXiv: Charting Gaps in Realistic Chart Understanding in Multimodal LLMs (arXiv:2406.18521v1) | 2 | 6% |
| Polyrating: A Cost-Effective and Bias-Aware Rating System for LLM Evaluation (arXiv:2409.00696v3) | 1 | 6% |
 LMArena leaderboard dataset | 1 | 6% |
| prior · pull toward the middle | 0% | |
| weighted average | = 42.9 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json