Mistral API Pricing (2026): Every Model, Batch Discount, and Cost Example

Mistral API Pricing (2026): Every Model, Batch Discount, and Cost Example

Mistral API pricing starts at $0.10 per 1M input tokens and $0.10 per 1M output tokens for Ministral 3 3B. Mistral Small 4 costs $0.15 input and $0.60 output, Mistral Medium 3.5 costs $1.50 / $7.50, and Mistral Large 4 is on sale at $0.68 / $2.09 (down from $1.36 / $4.18). Cached input is billed at 10% of the input rate, and the Batch API takes 50% off. All rates are from Mistral's official API pricing page, checked on October 20, 2026.

Below you'll find the full price list by model family, the free tier and batch rules, a cost calculator with worked examples, a comparison with GPT, Claude, Gemini, Qwen, and DeepSeek, and copy-paste setup code.

Mistral API pricing table (per 1M tokens)

Flagship and general models

Model	Input	Cached input	Output	Best for
Mistral Large 4 (sale price)	$0.68	$0.07	$2.09	Hard reasoning, coding, agents
Mistral Large 4 (standard)	$1.36	$0.14	$4.18	Price once the sale ends
Mistral Large 3	$0.50	$0.05	$1.50	General production work
Mistral Medium 3.5	$1.50	$0.15	$7.50	Strong general model, long context
Mistral Small 4	$0.15	$0.015	$0.60	High-volume chat, extraction, JSON
Ministral 3 14B	$0.20	$0.02	$0.20	Edge and low-latency tasks
Ministral 3 8B	$0.15	$0.015	$0.15	Classification, routing, tagging
Ministral 3 3B	$0.10	$0.01	$0.10	Cheapest text option

Code models

Model	Input	Cached input	Output
Codestral	$0.30	$0.03	$0.90
Codestral Embed	$0.15	$0.015	n/a

Specialized models

Model	Price
OCR 4.1	$4 per 1,000 pages ($0.40 cached)
Voxtral Mini Transcribe 2	$0.003 per minute
Voxtral TTS	$16 per 1M output characters
Mistral Moderation 2	Free
Z.ai GLM 5.3 (hosted by Mistral)	$1.40 input / $0.14 cached / $4.40 output

Source for all three tables: Mistral API pricing and Mistral Docs pricing, as of October 20, 2026. The pricing page also has Batch and Priority tabs and a regional-inference toggle, so confirm the tier you'll actually use.

Notable changes since earlier 2026 guides: Large 4's launch sale halves its standard price, Large 3 now costs less than Medium 3.5, and Mistral hosts a third-party model (GLM 5.3) on its own API. Older posts that still quote Mistral Large 2 at $2 / $6 are out of date.

Which Mistral model is the best value?

• Mistral Large 4 on sale is the standout deal. At $0.68 / $2.09, it costs less than Medium 3.5 ($1.50 / $7.50) on both input and output. While the sale lasts, test Large 4 before you default to Medium.
• Mistral Small 4 is the volume workhorse. At $0.60 output it is 12.5x cheaper than Medium on output tokens.
• Ministral 3 8B and 3B charge the same rate for input and output, which makes them unusually cheap for output-heavy tasks such as generating labels or short rewrites.
• Codestral at $0.30 / $0.90 suits code completion (FIM) workloads, where prompts are long and outputs are short.

Free tier, batch, and caching rules

Rule	What Mistral documents
Free usage	Mistral Studio accounts get a limited free mode with rate and usage limits. Production volume requires billing.
Batch API	50% discount on asynchronous jobs; up to 1M requests per batch; inline batching for under 10K requests (batch docs)
Batch endpoints	/v1/chat/completions, /v1/embeddings, /v1/fim/completions, /v1/ocr, /v1/moderations, /v1/classifications, /v1/conversations, /v1/audio/transcriptions
Cached input	Listed at 10% of the input rate for every text model on the pricing table
Regional inference	Separate regional rates via the toggle on the pricing page

Batch was introduced at "50% lower cost than that of a synchronous API call" in Mistral's batch API announcement. Each batch job runs a single model, so A/B testing two models means two jobs.

Mistral API cost calculator

cost = uncached_input/1e6 × input_rate
     + cached_input/1e6   × cached_rate
     + output/1e6         × output_rate
batch jobs: × 0.5

Worked examples (October 20, 2026 rates):

Workload (per month)	Model	Calculation	Cost
10M input + 2M output chatbot	Small 4	10 × $0.15 + 2 × $0.60	$2.70
Same chatbot	Medium 3.5	10 × $1.50 + 2 × $7.50	$30.00
Same chatbot	Large 4 (sale)	10 × $0.68 + 2 × $2.09	$10.98
500M-token document tagging, 20M output, batch	Ministral 3 8B	(500 × $0.15 + 20 × $0.15) × 0.5	$39.00
100K scanned pages	OCR 4.1	100 × $4	$400.00

The same calculator in Python:

MISTRAL = {  # USD per 1M tokens, mistral.ai/pricing/api, Oct 20, 2026
    "mistral-large-4-sale": (0.68, 0.07, 2.09),
    "mistral-medium-3.5":   (1.50, 0.15, 7.50),
    "mistral-small-4":      (0.15, 0.015, 0.60),
    "ministral-3-8b":       (0.15, 0.015, 0.15),
    "codestral":            (0.30, 0.03, 0.90),
}

def mistral_cost(model, uncached, cached=0, output=0, batch=False):
    i, c, o = MISTRAL[model]
    cost = (uncached * i + cached * c + output * o) / 1e6
    return cost * (0.5 if batch else 1)

print(mistral_cost("mistral-small-4", 10_000_000, output=2_000_000))  # 2.7

The table keys are labels for the calculator, not API model IDs. Get the exact IDs from GET /v1/models or the model overview.

Mistral vs other LLM APIs: price per 1M tokens

Mistral rates are from mistral.ai. The other rows use the list prices on each model's Flatkey model page. All figures were checked on October 20, 2026.

Model	Input	Output	Tier
Mistral Small 4	$0.15	$0.60	Budget
qwen3.8-flash	$0.15	$0.47	Budget
claude-haiku-5-5	$0.10	$0.50	Budget
gpt-6-luna	$0.10	$0.50	Budget
Mistral Large 4 (sale)	$0.68	$2.09	Frontier
qwen3.8-max	$2.00	$6.00	Frontier
gpt-6-sol	$2.00	$10.00	Frontier
claude-sonnet-5-5	$2.00	$10.00	Frontier

Is Mistral cheaper than OpenAI? At the frontier tier, yes, by a wide margin while the Large 4 sale runs: $2.09 output against $10.00 for gpt-6-sol. At the budget tier, Small 4's $0.60 output is slightly above the $0.50 list output of GPT-6 Luna and Claude Haiku 5.5. Price alone doesn't settle the choice, so compare accepted-answer rates on your own evals. For more providers, see our cheapest LLM APIs comparison.

How to get a Mistral API key and make a first request

1. Sign in to Mistral Studio (quickstart) and open API Keys.
2. Create a key, set an expiry, and copy it. The full key is shown only once.
3. Export it as MISTRAL_API_KEY and call https://api.mistral.ai/v1.

curl https://api.mistral.ai/v1/chat/completions \
  -H "Authorization: Bearer $MISTRAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "mistral-small-latest",
       "messages": [{"role": "user", "content": "Return one sentence about Paris."}]}'

import os
from mistralai.client import Mistral   # pip install mistralai

client = Mistral(api_key=os.environ["MISTRAL_API_KEY"])
r = client.chat.complete(
    model="mistral-small-latest",
    messages=[{"role": "user", "content": "Extract company and amount as JSON: ACME paid $40."}],
)
print(r.choices[0].message.content)

Aliases such as mistral-small-latest move to new versions automatically. Pin a dated model ID from GET /v1/models when you need reproducible output. See the Mistral API docs for the full endpoint reference.

Using Mistral alongside other models with one key

As of October 20, 2026, the Flatkey model directory doesn't list Mistral models, so call Mistral directly. A common pattern is to keep Mistral on its own client and use one OpenAI-compatible Flatkey key (https://router.flatkey.ai/v1) for GPT, Claude, Gemini, Qwen, DeepSeek, Kimi, and GLM. Your code then needs only two clients instead of six.

import os
from openai import OpenAI

mistral = OpenAI(api_key=os.environ["MISTRAL_API_KEY"], base_url="https://api.mistral.ai/v1")
flatkey = OpenAI(api_key=os.environ["FLATKEY_API_KEY"], base_url="https://router.flatkey.ai/v1")

def run(prompt, model):
    client = mistral if model.startswith(("mistral", "ministral", "codestral")) else flatkey
    r = client.chat.completions.create(model=model, messages=[{"role": "user", "content": prompt}])
    return r.choices[0].message.content, r.usage

for m in ["mistral-small-latest", "qwen3.8-flash", "claude-haiku-5-5"]:
    text, usage = run("Classify sentiment: 'Shipping was slow but support was great.'", m)
    print(m, usage.prompt_tokens, usage.completion_tokens, text[:60])

Mistral's chat endpoint accepts OpenAI-style requests for common chat calls. Test tool calls and structured output on your own prompts before you rely on them. The same switch works from Node or curl:

import OpenAI from "openai";
const flatkey = new OpenAI({ apiKey: process.env.FLATKEY_API_KEY, baseURL: "https://router.flatkey.ai/v1" });
const r = await flatkey.chat.completions.create({
  model: "gpt-6-luna",
  messages: [{ role: "user", content: "Summarize in 10 words: batch APIs halve cost." }],
});
console.log(r.choices[0].message.content);

curl https://router.flatkey.ai/v1/chat/completions \
  -H "Authorization: Bearer $FLATKEY_API_KEY" -H "Content-Type: application/json" \
  -d '{"model": "claude-haiku-5-5", "messages": [{"role": "user", "content": "ping"}]}'

More setup detail is in the Flatkey quickstart and the OpenAI API base URL guide.

Common Mistral API errors

Status	Likely cause	Fix
401	Missing or expired key	Check MISTRAL_API_KEY and the workspace
402	Billing not active for the requested usage	Add a payment method or leave Free mode
404	Wrong path (for example /v1/chat/completions set as the base URL)	Use https://api.mistral.ai/v1 as the base URL
422	Invalid parameter for the model	Remove unsupported fields and check the model card
429	Rate or spend limit reached	Back off exponentially and review workspace limits (rate-limit guide)

FAQ

How much does the Mistral API cost?

As of October 20, 2026, prices range from $0.10 / $0.10 per 1M tokens (Ministral 3 3B) to $1.50 / $7.50 (Medium 3.5). Small 4 costs $0.15 / $0.60, and Large 4 is on sale at $0.68 / $2.09.

Is the Mistral API free?

Mistral Studio offers a limited free mode with rate and usage limits. Mistral Moderation 2 is listed as free. Beyond that, usage is billed per token.

What is the cheapest Mistral model?

Ministral 3 3B, at $0.10 per 1M input and output tokens. For a more capable general model, Mistral Small 4 costs $0.15 input and $0.60 output.

Is Mistral better than GPT?

It depends on the task. Mistral Large 4's sale price is far below GPT frontier list prices, but quality differs by workload. Run the same eval set on both and compare cost per accepted answer.

Does Mistral offer a batch discount?

Yes. The Batch API is 50% cheaper than synchronous calls and supports up to 1 million requests per job.

How much does Mistral OCR cost?

OCR 4.1 costs $4 per 1,000 pages, or $0.40 per 1,000 for cached pages, as of October 20, 2026.

Bottom line

Start with Mistral Small 4 ($0.15 / $0.60) as your baseline. While the sale lasts, test Large 4 at $0.68 / $2.09 before paying more for Medium 3.5, and push offline work through the 50%-off Batch API. Re-check Mistral's pricing page before you lock a budget. To compare Mistral against other model families, use the Flatkey model directory and pricing page.

Last updated: October 20, 2026 · Flatkey Team. Prices are in USD and quoted from the official sources linked inline.