Skip to main content
Watch: I Built an AI App Without Leaving Databricks

Cost levers - verified

How to cut your Mistral API bill

Prompt caching: automatic. A cache hit typically saves 90% on input across 8 current Mistral AI models. Batch: 50% off, 24-hour period. Batch and caching stack: not stated. Service tiers: Priority Tier (1.75x).

Read from docs.mistral.ai, mistral.ai. Compare every provider's cost levers

Caching
Automatic
Cache hit saves
90%
Batch
50% off
Service tiers
1

Prompt caching

Mode

Automatic: repeated prefixes are cached with no code change

Prefix caching; prompt_cache_key raises (but does not guarantee) hit rates.

Cache lifetime
Not stated
Minimum prompt

64 tokens (cache blocks of 64 tokens; prompts under 64 tokens get no hits)

Cache write cost

No separate cache-write charge described; uncached tokens billed at the standard input price

Cache read cost

10% of standard input price

Storage cost
Not stated
How to enable

Set the same prompt_cache_key on requests likely to share a prefix.

Request fragment
"prompt_cache_key": "support-session-42"
From the docs
“Cached prompt tokens are billed at 10% of the standard input token price, and cache hits can reduce response latency by reusing part of the previous prompt computation.”
“`prompt_cache_key` increases the chance of a cache hit, but it doesn't guarantee one.”
“Prompts with fewer than 64 prompt tokens do not have cache hits.”
“Cache blocks contain 64 tokens.”
“`5` uncached input tokens billed at the standard input token price.”
“cached input tokens reduce input cost by up to 90% for repeated prompts.”

Sources: docs.mistral.ai/studio/conversations/advanced/prompt-caching, mistral.ai/pricing

Batch API

Discount

50% off

50% discount

Window

Requests are queued and processed over a 24-hour period

Limits

Up to 1 million requests per batch (file batching); inline batching for fewer than 10,000 requests; one model per batch

Endpoint

Batch jobs (Studio or API) for /v1/chat/completions, /v1/embeddings, /v1/fim/completions, /v1/moderations, /v1/ocr, /v1/conversations, /v1/audio/transcriptions and more

Stacks with caching

Not stated: the provider's pages do not say

From the docs
“Batching allows you to run asynchronous inference on large inputs in parallel, reducing compute costs while running large workloads at a 50% discount.”
“| Bursting | Requests are queued and processed over a 24-hour period | Standard rate-limit behavior | Falls back to Standard Tier when Priority Tier limits are exceeded |”
“We support up to 1 million requests in a single batch”
“For batches of fewer than 10,000 requests, we support inline batching.”

Sources: docs.mistral.ai/studio/batch-processing, docs.mistral.ai/inference/priority-tier, mistral.ai/pricing

Service tiers

Flex trades speed or availability for a lower price; Priority and Fast cost more for lower latency or reserved capacity.

Priority

Priority Tier

Set with service_tier: auto (default standard_only)

Price effect

1.75x standard list pricing (75% premium on input, output and cached tokens), applied after the cache discount

Trade-off

Priority queue (seconds), 99.5% uptime SLA, falls back to Standard when limits exceeded; requires account setup with Mistral

From the docs
“Priority Tier is billed as a premium multiplier on top of Standard Tier list pricing. The multiplier is 1.75x standard list pricing, which is a 75% premium on input, output, and cached tokens.”
docs.mistral.ai
“Priority Tier requires account setup with Mistral.”
docs.mistral.ai
“| Uptime SLA | Not available | Not available | 99.5% |”
docs.mistral.ai

Other discounts and surcharges

Other discounts on the pages checked: Not stated

Mistral AI per-model prices

USD per 1M tokens from the provider's own pricing page, newest release first. A dash means the model has no such rate. These are the Mistral AI models billed through Mistral. Muted rows are scheduled for shutdown or deprecated.

Mistral AI models with input, cache, batch, output and fast mode prices per 1M tokens
ModelInputCached inputBatch inputOutputBatch output
Mistral Large 4$1.36$0.1489.7% off$0.68$4.18$2.09
Mistral Medium 3.5$1.50$0.1590% off$0.75$7.50$3.75
Mistral Small 4$0.15$0.01590% off$0.075$0.60$0.30
Ministral 3 14B$0.20$0.0290% off$0.10$0.20$0.10
Ministral 3 3B$0.10$0.0190% off$0.05$0.10$0.05
Ministral 3 8B$0.15$0.01590% off$0.075$0.15$0.075
Mistral Large 3$0.50$0.0590% off$0.25$1.50$0.75
Codestral 25.08$0.30$0.0390% off$0.15$0.90$0.45

Research notes

Batch page FAQ answers (max jobs, expiry) are collapsed accordions that did not render; only the inline limits above were verified. TTL not documented.

FAQ

Does Mistral cache prompts automatically?

Automatic: repeated prefixes are cached with no code change. Prefix caching; prompt_cache_key raises (but does not guarantee) hit rates. How to enable: Set the same prompt_cache_key on requests likely to share a prefix.

How much does a cache hit cost on Mistral?

Across 8 current models with a cache read price, a hit saves 90% (89.7% to 90%) off the input price. The docs say: 10% of standard input price.

Does Mistral have a batch API discount?

50% discount. Window: Requests are queued and processed over a 24-hour period. Stacks with caching: not stated: the provider's pages do not say.

What service tiers does Mistral offer?

Priority Tier: 1.75x standard list pricing (75% premium on input, output and cached tokens), applied after the cache discount.

Sources

Last verified from Mistral AI's own pages below. Quotes are verbatim; each was matched against the live page.

Newsletter

Know when a provider changes its prices

One email per week. New model releases, pricing changes and what they mean for what you build.