StackLift
Mistral · updated Nov 22, 2026, 4:07 PM UTC

Saba API pricing.

Saba by Mistral costs $0.20 per million input tokens and $0.60 per million output tokens, with a 33K-token context window. That puts it in the budget range of the models we track. A support chatbot handling 10,000 conversations a month would cost about $6.00 a month in API fees.

Price

What Saba costs per token

List prices, checked every hour. Blended assumes three input tokens for every output token. This model sits in the budget range of the models we track.

ModelProviderInput / 1MOutput / 1MBlended / 1MContext
SabaMistral$0.20$0.60$0.3033K

Source: Saba in the OpenRouter model catalog. Check it against Mistral AI's pricing.

Specs

What Saba can do

From Saba's entry in the OpenRouter model catalog.

Reads
text and files (PDFs and documents)
Writes
text
Context window
33K tokens
Longest answer
26K tokens
Tool calling
Yes: it can call functions and APIs you define
Structured output
Yes: answers can follow a JSON schema
Reasoning
Not listed
Cached input
$0.02 per 1M tokens (90% off the input price)
Open weights
No: API access only
Knowledge cutoff
September 2024

Monthly cost

What common workloads cost on Saba

API fees only, at list price. The assumptions are shown, so you can scale them to your own volume.

Customer support chatbot

$6.00 / month

10,000 conversations a month, ~1,500 in and ~500 out each: 15M input and 5M output tokens.

Knowledge-base Q&A (RAG)

$20.80 / month

20,000 questions a month, ~4,000 in and ~400 out each: 80M input and 8M output tokens.

Document summaries and extraction

$5.76 / month

2,000 documents a month, ~12,000 in and ~800 out each: 24M input and 1.6M output tokens.

Coding assistant for a team of 10

$9.60 / month

10 developers a month, 3.0M in and 0.6M out each: 30M input and 6M output tokens.

Hosting, a vector database, monitoring and the build itself are extra. The free StackLift estimate adds them for your exact scope.

Per request

What one request costs on Saba

A request costs its input tokens times the input price plus its output tokens times the output price. A typical assistant turn of 2,000 input and 500 output tokens on Saba: 2,000 × $0.20 ÷ 1M + 500 × $0.60 ÷ 1M = $0.0007.

RequestInput / output tokensOne request1,000 requests
Short chat reply500 / 200$0.0002$0.22
Typical assistant turn2,000 / 500$0.0007$0.70
Long document task20,000 / 1,500$0.0049$4.90

As you grow

Saba cost at ten times less and ten times more

The same workloads at a tenth and at ten times the volume, at Saba's list price.

WorkloadA tenthTypicalTen times
Customer support chatbot$0.601,000 conversations$6.0010,000 conversations$60.00100,000 conversations
Knowledge-base Q&A (RAG)$2.082,000 questions$20.8020,000 questions$208200,000 questions
Document summaries and extraction$0.58200 documents$5.762,000 documents$57.6020,000 documents

Fit

When Saba is the right choice

Saba is priced for volume: classification, extraction, routing and short replies, where thousands of requests a day must stay cheap. Hard reasoning may need a larger model for some steps.

Its 33K-token context window suits short requests; long documents need retrieval (RAG) or splitting.

Saba reads files (PDFs and documents) as well as text and can call tools and APIs and return structured JSON, so it can drive agents that act in your systems.

Many products mix models: a cheaper one for simple steps and a stronger one where quality matters. Test a shortlist on your own examples before you commit.

Save

Ways to pay less for Saba

  1. 1. Cache repeated instructions

    Saba charges $0.02 per million cached input tokens instead of $0.20. A 4,000-token system prompt sent with 10,000 requests a month costs $8.00 at the list price and about $0.80 when cached.

  2. 2. Route by difficulty

    Mistral Small 3.2 24B costs 65% less per token than Saba. Send simple requests there and keep Saba for the steps that need it.

  3. 3. Batch what can wait

    Overnight or bulk jobs often qualify for batch pricing, at a lower rate than live requests.

  4. 4. Ask for shorter answers

    Output costs 3 times as much as input on Saba, so a tight answer format (bullet points, JSON, a word limit) cuts the bill more than a shorter prompt.

Alternatives

Models priced closest to Saba

From other providers, nearest on blended price. Quality and speed differ by task: test the shortlist on your own examples.

ModelProviderInput / 1MOutput / 1MBlended / 1MContext
SabaMistral$0.20$0.60$0.3033K
DeepSeek V3.2DeepSeek$0.26$0.38$0.29164K
Qwen3.8 Omni FlashQwen$0.15$0.47$0.231M
Qwen3.8 FlashQwen$0.15$0.47$0.231M
Qwen3.8 27BQwen$0.035$1.50$0.4011M
DeepSeek V3.1DeepSeek$0.25$0.95$0.425164K

Mistral

Other Mistral models

The same provider's newest models, for a quicker or more capable option.

ModelProviderInput / 1MOutput / 1MBlended / 1MContext
Mistral Large 4Mistral$0.70$2.10$1.051M
Mistral Medium 3.5Mistral$1.50$7.50$3.00262K
Mistral Medium 3.1Mistral$0.44$2.20$0.88131K

Questions about Saba pricing

How much does Saba cost per 1,000 tokens?

$0.0002 per 1,000 input tokens and $0.0006 per 1,000 output tokens, from the list price of $0.20 and $0.60 per million.

Is Saba cheaper than DeepSeek V3.2?

No, on list price. Saba averages $0.30 per million tokens (3 parts input to 1 output) against $0.29 for DeepSeek V3.2. A pricier model can still cost less overall if it needs fewer retries or shorter prompts.

What is Saba's context window?

33K tokens: the most text the model reads and writes in one request. Long documents beyond that are split, or retrieved in parts with RAG.

How much would a chatbot cost on Saba?

About $6.00 a month for 10,000 conversations a month, ~1,500 in and ~500 out each. Caching repeated instructions and routing simple questions to a cheaper model lower it further.

Does Saba support tool calling?

Yes. Saba can call functions and APIs you describe to it, which is what lets an AI agent look up data or take actions. It can also return answers that follow a JSON schema.

Can Saba read images or documents?

Yes. Besides text, Saba reads files (PDFs and documents).

How long can Saba's answers be?

Up to 26K tokens per answer (about 20,000 words), within a 33K tokens context window.

Does Saba have prompt caching?

Yes. Input read from the cache costs $0.02 per million tokens, 90% less than the $0.20 list price, which suits long system prompts and shared context.

How do I estimate the AI cost of my own product?

Count the requests a month and the tokens in and out of each, then multiply by the prices here. StackLift's free estimate does this for your exact scope, alongside the cost to build it.

Build with it

Scope a product on Saba

Pick your features and integrations; the estimate returns the build cost, timeline and team, and the model's monthly running cost. Free, no signup.