Reka Flash 3
Reka Flash 3 by Reka AI is a 21B reasoning model with 66K context, explicit reasoning tags, budget forcing, and function calling for compact deployments.
Model Overview
Capabilities, design details, and architectural traits
Reka Flash 3 - Compact reasoning model
Reka Flash 3 is Reka’s 21B-parameter reasoning model for compact deployments, with documentation emphasizing explicit reasoning traces, function calling, and a long but practical context window.
Defining Characteristics
- 21B parameter reasoning model: Reka describes Flash 3 as a 21-billion-parameter model trained from scratch for general-purpose reasoning.
- Explicit reasoning tags: The model uses
<reasoning>tags to expose internal thought traces in supported integrations. - Budget forcing for latency control: Provider documentation says Flash 3 supports a budget forcing mechanism to cap reasoning steps and control latency.
- 66K context window: Public listings document a 65,536/66K-token context window and matching maximum output size.
Core Design
- The model is text-first and English-focused in the official model card, with on-device or low-latency deployment called out as a key use case.
- Reka also positions Flash 3 as suitable for function calling, coding, and instruction following.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Reka AI | Anthropic | Anthropic |
| Release Date | March 10, 2025 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | Jan 2025 | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 32K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $0.20 Best Input Pricing | $4 | $2 |
| Output Pricing | $0.80 Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | text | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 5.6 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | - | - | - |
| Agentic Index | - | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Top AI Models
Leading alternatives by intelligence score
