Mistral vs Claude: EU regions and price 2026

TL;DR

Mistral vs Claude in the EU is a residency choice as much as a price choice. Claude Sonnet 5.5 costs 2 / 10 USD per million tokens and stays on Bedrock behind the EU geo profile. Mistral Large 3 costs 0.50 / 1.50 USD in US regions and has no EU region on Bedrock. For code in Frankfurt use Devstral 2 at 0.48 / 2.40. For Large 3 in the EU use Foundry Data Zone or self-host.

What is the difference between Mistral and Claude for a company?

Mistral vs Claude in 2026 is a three-line comparison, not a chatbot bake-off. Claude ships Sonnet 5.5 and Opus 5.5 with a 1 million token context and an EU geo profile on Bedrock. Mistral ships Large 3 as an open MoE with 675 billion parameters, but Bedrock lists no EU region for it. Devstral 2 covers code in Frankfurt. Pixtral Large covers images on an EU profile.

The Sonnet 5.5 card of 28 September 2026 lists a 1 million token context, 128,000 output tokens and a June 2026 knowledge cutoff. The EU ID is eu.anthropic.claude-sonnet-5-5. Anthropic treats Sonnet 5.5 as the production model and Opus 5.5 as the coding and knowledge-work model. Which Claude to pin first is in Claude Sonnet vs Opus.

The Mistral Large 3 card lists 675 billion parameters, a 2 December 2025 launch, an EOL no sooner than 2 December 2026, 256,000 context tokens and 32,000 output tokens. The ID is mistral.mistral-large-3-675b-instruct. Geo and global profiles are not supported. The Hugging Face card describes the same model as a mixture of experts with 41 billion active parameters, an Apache 2.0 licence and a vision encoder.

Feature	Claude Sonnet 5.5	Mistral Large 3	Devstral 2 123B	Pixtral Large
Role	Production, agents	Frontier, multilingual, vision	Software engineering	Text and image
Launch	28 September 2026	2 December 2025	June 2025	19 November 2024
Context / output	1M / 128K	256K / 32K	256K / 32K	128K / 16K
Bedrock ID	eu.anthropic.claude-sonnet-5-5	mistral.mistral-large-3-675b-instruct	mistral.devstral-2-123b	eu.mistral.pixtral-large-2502-v1:0
EU on Bedrock	Geo profile	No EU region	In-region Frankfurt, Stockholm, Ireland, Milan, London	Geo profile
Weights	Closed	Apache 2.0	Closed on Bedrock	Closed on Bedrock

Pixtral Large is the only Mistral model with an eu. profile. The Pixtral card lists 124 billion parameters. Devstral 2 is a 123 billion parameter coding model on the Devstral card, lifecycle Legacy, model EOL 30 March 2027. We would not set Devstral 2 as a new default for 2027.

How much does Mistral vs Claude cost per task?

Mistral Large 3 is four times cheaper on input than Claude Sonnet 5.5 on Bedrock in the US. The Bedrock pricing page lists Large 3 in US East and West at 0.50 / 1.50 USD per million tokens. Anthropic lists Sonnet 5.5 at 2 / 10 USD, with a 10 percent premium on regional endpoints. In the EU you therefore pay 2.20 / 11 USD for Claude and you cannot buy Large 3 on Bedrock at all.

Token prices read on 13 October 2026, standard on-demand, USD per million tokens:

Route	Input / output	Cache read	Batch
Claude Sonnet 5.5, Claude API global	2.00 / 10.00	0.10	1.00 / 5.00
Claude Sonnet 5.5, Bedrock EU or Google Europe	2.20 / 11.00	0.22 (10 percent rule)	50 percent of the EU row
Claude Opus 5.5, Claude API global	4.00 / 20.00	0.20	2.00 / 10.00
Mistral Large 3, Bedrock US East/West	0.50 / 1.50	not published here	not published here
Devstral 2, Frankfurt / Stockholm / Ireland / Milan	0.48 / 2.40	not published here	not published here
Devstral 2, London	0.62 / 3.10	not published here	not published here

Pixtral Large is missing from the current pricing page. We do not invent a dollar rate. Foundry lists Mistral-Large-3 in Data Zone Standard without a public dollar tariff on the region page. The full Claude table is in Claude API pricing in the EU.

The ticket rows below are our arithmetic, not a vendor quote. Assumption: 3,000 input and 400 output tokens, 10,000 jobs a month.

Workload	Sonnet 5.5 global	Sonnet 5.5 EU (+10 %)	Large 3 Bedrock US	Devstral 2 Frankfurt
10,000 support replies	100 USD	110 USD	21 USD	24 USD
1,000 reviews at 40,000 / 2,000 tokens	100 USD	110 USD	23 USD	24 USD

Large 3 wins on price only if US inference is allowed. When the DPO requires EU processing, that US row disappears. Then you keep Devstral 2 for code, Claude on the EU profile for everything else, or Large 3 on Foundry Data Zone or on your own GPUs. Another open-weight price path is in DeepSeek pricing on Bedrock EU.

Does inference stay in the EU?

Claude Sonnet 5.5 stays in the EU if you call eu.anthropic.claude-sonnet-5-5. Mistral Large 3 does not stay in the EU on Bedrock: the region list names us-east-1, us-east-2, us-west-2, Tokyo, Mumbai, Sydney and Sao Paulo. Foundry Data Zone Standard lists Large 3 in France Central, Germany West Central, Italy North, Poland Central, Spain Central, Sweden Central and West Europe.

From Frankfurt the Claude EU profile routes to Frankfurt, Stockholm, Milan, Spain, Ireland and Paris. London as a source can also hit London. That is Europe, not a pin to one data centre. How Bedrock geo profiles work is in the Amazon Bedrock EU guide.

Devstral 2 is in-region. The pricing page names Europe (Ireland), Europe (Milan), Europe (Frankfurt), Europe (Stockholm) and Europe (London). London is not in the EU. For a processing record that says EU member states, use Frankfurt, Stockholm, Ireland or Milan, not London.

Pixtral Large keeps destinations in Frankfurt, Stockholm, Ireland and Paris through eu.mistral.pixtral-large-2502-v1:0. There is no global profile.

Google Cloud lists Claude Sonnet 5.5 on the Europe multi-region at 1,250 requests per minute, 12,500,000 uncached input TPM and 1,250,000 output TPM. Mistral Large 3 is not on that catalogue. Mistral Medium 3 sits in europe-west4 at 90 requests per minute, 315,000 TPM and a 128,000 token context. Medium 3 is not Large 3.

Foundry Global Standard may process prompts in any region where the model is deployed. Data Zone Standard keeps processing inside the EU data zone. The region page states that. Claude still has no EU data zone on Foundry in the comparison we keep in the Foundry guide and the GDPR provider comparison. Mistral Large 3 does. That flips the usual cloud pick: Claude on Bedrock or Google, Large 3 on Foundry Data Zone or self-hosted.

When should you pick Mistral, and when Claude?

Pick Claude Sonnet 5.5 on Bedrock EU when prompts must stay in member states and you need agents, computer use or a 1 million token context. Pick Devstral 2 in Frankfurt when the job is code and the bill should sit under Sonnet. Pick Mistral Large 3 only when Foundry Data Zone, self-hosting or a US Bedrock region fits the processor record.

1. Write residency into the brief: one member state, EU geography, or global.
2. Score the task on a fixed eval pack, not a demo chat.
3. Pin IDs: Claude only through eu.anthropic.claude-sonnet-5-5 or Google region="eu". Devstral only as mistral.devstral-2-123b in eu-central-1. Large 3 on Bedrock only with an explicit US region.
4. Put a spend cap on a LiteLLM gateway so apps cannot silently flip to global. or us-east-1.
5. Treat Devstral 2 as time-boxed. Model EOL is 30 March 2027. Name the successor before procurement freezes on Legacy.
6. If weights must sit in your data centre, host Large 3 yourself. The Hugging Face card names FP8 on one B200 or H200 node and vllm >= 1.12.0 with --tensor-parallel-size 8.

We recommend Bedrock EU for Claude as the default in European companies. Mistral Large 3 wins on price and licence, not on AWS residency. If you want open weights and EU processing together, use Foundry Data Zone or your own GPUs, not the bare Bedrock ID.

How do you call both models on Bedrock?

Two Converse calls pin Claude on the EU profile and Devstral in Frankfurt. We leave Large 3 in us-east-1 on purpose so the region break is visible. Use AWS_BEARER_TOKEN_BEDROCK or IAM, not an Anthropic key.

import boto3

def complete(region, model_id, prompt):
    client = boto3.client("bedrock-runtime", region_name=region)
    response = client.converse(
        modelId=model_id,
        messages=[{"role": "user", "content": [{"text": prompt}]}],
        inferenceConfig={"maxTokens": 512},
    )
    text = "".join(
        block["text"] for block in response["output"]["message"]["content"] if "text" in block
    )
    return {"model": model_id, "region": region, "text": text, "usage": response.get("usage")}

prompt = "Summarise the ticket in four bullets."
claude = complete("eu-central-1", "eu.anthropic.claude-sonnet-5-5", prompt)
devstral = complete("eu-central-1", "mistral.devstral-2-123b", prompt)
# Large 3 has no EU region. Call it only if the DPA allows US processing.
large = complete("us-east-1", "mistral.mistral-large-3-675b-instruct", prompt)
print(claude["model"], devstral["model"], large["region"])

1. Create the client in eu-central-1 when inference should start in Frankfurt.
2. Pass the eu. profile for Claude, not global.anthropic.claude-sonnet-5-5.
3. Pass the bare Devstral ID in an EU region from the pricing page.
4. Log modelId and region next to the request ID, or FinOps only sees one Bedrock account.
5. For Large 3 in the EU, change provider: Foundry Data Zone or vLLM on your hardware, not Bedrock.

Self-hosting follows the Hugging Face card: vllm serve mistralai/Mistral-Large-3-675B-Instruct-2512 --max-model-len 262144 --tensor-parallel-size 8. That wants an 8-way H200 node. We have not read EU rental rates and do not quote them.

FAQ

How much does Mistral vs Claude cost for 10,000 support tickets?

On our assumptions (3,000 input, 400 output) Claude Sonnet 5.5 costs 100 USD globally and 110 USD in the EU. Mistral Large 3 on Bedrock in the US costs 21 USD. Devstral 2 in Frankfurt costs 24 USD. Large 3 is the cheap row only when US processing is allowed.

Mistral vs Claude: which model should a European company default to?

Claude Sonnet 5.5 through eu.anthropic.claude-sonnet-5-5 when personal prompts must stay in the EU. Devstral 2 in Frankfurt when the job is code and the price should sit under Sonnet. Mistral Large 3 first only on Foundry Data Zone, self-hosting, or a DPA that allows US Bedrock.

Does Mistral Large 3 stay in the EU on Amazon Bedrock?

No. The model card and the region list read on 13 October 2026 name no EU region and no geo profile. The listed regions are US East, US West, Tokyo, Mumbai, Sydney and Sao Paulo. For EU processing of Large 3 use Foundry Data Zone or your own hardware.

Should we self-host Mistral Large 3 or run it on Foundry?

Self-host when the weights must sit in your data centre. The card names Apache 2.0, FP8 on a B200 or H200 node, and vLLM 1.12.0 or newer. Use Foundry Data Zone when you will not run GPUs and you want processing in France Central, Germany West Central, Italy North, Poland Central, Spain Central, Sweden Central or West Europe. We did not read a dollar rate on the region page.

Claude on Google or Mistral Medium 3 on Google?

Claude Sonnet 5.5 on the Europe multi-region has 1,250 QPM and a 1 million token context. Mistral Medium 3 in europe-west4 has 90 QPM, 315,000 TPM and a 128,000 token context. Medium 3 is not Large 3. The frontier comparison is missing Large 3 on Google.

Does Pixtral Large have an EU price on Bedrock?

On 13 October 2026 Pixtral Large was absent from the Bedrock pricing page. The EU profile and destination regions are documented. We leave the token price blank until AWS publishes it again.

Sources

1. Anthropic: Pricing (13 October 2026)
2. AWS: Claude Sonnet 5.5 model card (13 October 2026)
3. AWS: Mistral Large 3 model card (13 October 2026)
4. AWS: Devstral 2 123B model card (13 October 2026)
5. AWS: Pixtral Large model card (13 October 2026)
6. AWS: Regional availability by models (13 October 2026)
7. Amazon Bedrock Pricing (13 October 2026)
8. Hugging Face: Mistral-Large-3-675B-Instruct-2512 (13 October 2026)
9. Microsoft Learn: Foundry region availability (13 October 2026)
10. Google Cloud: Mistral Medium 3 (13 October 2026)
11. Google Cloud: Claude Sonnet 5.5 (13 October 2026)