Last verified
MID64K CONTEXTREASONING BILLED AS OUTPUTAPACHE-2.0 WEIGHTS

Reka Flash 3 API Pricing

Reka Flash 3 (reka-flash-3) is Reka's 21B reasoning model for chat, coding and instruction-following, and since Reka Core and Reka Flash left the catalog it is the company's only general-purpose text model. Reka lists $0.1/M input and $0.2/M output, with a 64K context and no cache rate. The thinking trace is billed as output.

Input - per 1M tokens
$0.10/M
Source Reka flat
Output - per 1M tokens
$0.20/M
Context 64K incl. reasoning
Cached input - per 1M tokens
Not published
Cache dash on Reka pricing no cache rate
Effective - agentic blend
$0.108/M
92/8 split - no cache discount
§ 01 / TERMINAL

Reka Flash 3 API cost calculator.

Live calculator pre-loaded with current Reka Flash 3 token rates. Count the reasoning trace in the output share: on this model it is billed as output, so a chat-style 70/30 mix is closer to real use than the agentic default.

$ /mo
Workload split
Tokens you can process
—
Words equivalent (English)
—
Effective rate
—
Open full calculator (all models · share URL · CSV) →
§ 02 / SCENARIOS

Reka Flash 3 cost on real workloads.

CHATBOT

Assistant turn with reasoning

$0.0008/turn
3,000 in / 2,500 out~125,000 turns/$100
CODING

Function fix

$0.002/fix
8,000 in / 6,000 out~50,000 fixes/$100
RAG

Grounded answer

$0.0016/answer
12,000 in / 2,000 out~62,500 answers/$100
DOC

Long document summary

$0.0058/doc
50,000 in / 4,000 out~17,241 docs/$100
§ 03 / TOKENIZER

Reka Flash 3 token counter and cost.

What's a token →
Estimate · reka-tokenizer-estimate · ≈3.85 chars/token Auto-counts as you type

This is a chars-per-token approximation, not a real tokenizer. Actual tokens vary by language, code density, and tool-call overhead — counts are typically ±10–20% off for English prose, more for code or non-Latin scripts. For exact billing, use the vendor's official tokenizer.

Characters 401
Words 69
Tokens (estimated) 104 tokens
Cost as input · uncached $0.00001 USD
Cost as output · uncached $0.00002 USD
§ 04 / SHELF

Reka Flash 3 vs similar models.

All models →
Model Input /M Output /M Effective blended Context Best for
Reka Edge $0.10 $0.10 $0.10 −7% 16K Image and video input on a 16K window
Step 3.5 Flash $0.10 cache $0.02 $0.30 $0.0556 −48% 256K Cheap reasoning with a 256K window
Solar Mini 4 $0.10 cache $0.01 $0.40 $0.0561 −48% 512K Long-context work at a similar input rate
GPT-6 Luna $0.10 cache $0.01 $0.50 $0.0682 −37% 1.05M Low-cost routing with image input
Command R7B $0.0375 $0.15 $0.0465 −2.3× 128K Lowest per-token cost for simple chat

Reka Flash 3 API pricing FAQ.

Practical Reka Flash 3 pricing questions: what the two rates buy, why reasoning changes the bill, and how it relates to the Reka models that left the catalog.

Q · 01 How much does Reka Flash 3 cost? +
Reka's model catalog and pricing page list reka-flash-3 at $0.1/M input and $0.2/M output, read October 29, 2026. Neither page publishes a cache-read rate, a batch discount or a context-length surcharge for it.
Q · 02 Why is my output bill higher than the answers I see? +
Because Reka Flash 3 is a reasoning model and the catalog says the trace is billed as output. Every response carries thinking tokens before the visible answer, so output is usually a larger share of the bill than on a plain chat model. The scenarios on this page budget for that; the agentic 92/8 tile assumes a short-answer mix and understates reasoning-heavy chat.
Q · 03 Is prompt caching priced separately? +
No. Reka's pricing table prints a dash in the cache-read column for this model, while it does print one for some of the third-party models it hosts. AI//COST keeps cached input at the full input rate, $0.1/M, rather than assuming a discount.
Q · 04 What happened to Reka Core and Reka Flash? +
Both left Reka's catalog when it moved its prices from docs.reka.ai to developer.reka.ai between August and October 2026, without an announced shutdown date. Their pages keep the last prices Reka published. Reka names no successor, so Reka Core and Reka Flash point here: Flash 3 is the only general-purpose Reka text model left.
Q · 05 Can I run Reka Flash 3 myself? +
Yes. Reka published the weights under Apache-2.0 at huggingface.co/RekaAI/reka-flash-3 in March 2025, in a Llama-compatible format that common inference servers load. Self-hosting removes the per-token price and replaces it with your own GPU cost, which this page does not estimate.
Q · 06 Which API model ID should I use, and in what language? +
Use reka-flash-3, the exact string Reka's catalog gives for the model field of its OpenAI-compatible chat completions endpoint. Reka describes the model as primarily English with limited multilingual understanding, so for non-English work compare it with a multilingual model before committing.
Y
Reviewed by Yaroslav Vikhariev Founder · AI//COST · Prices checked by hand against developer.reka.ai · Last verified Oct 29, 2026
Methodology Report a correction More by Y.V.