Llama vs Claude: EU regions, EOL 2026

TL;DR

Llama vs Claude in the EU is a residency and lifecycle split. Claude Sonnet 5.5 costs 2 / 10 USD per million tokens plus 10 percent on regional endpoints and stays behind eu.anthropic.claude-sonnet-5-5. The only Llama EU profiles on Bedrock, 1B and 3B, reached model EOL on 7 July 2026. Llama 4 Scout and Maverick expose a US geo ID only.

What separates Llama and Claude for companies?

Llama vs Claude in 2026 is two product lines with opposite residency models. Claude Sonnet 5.5 gives 1 million tokens of context behind an active EU geo profile on Bedrock. Llama 4 ships open weights under the Community License and stays on Bedrock behind us.meta. IDs. The only Llama EU profiles are Legacy, with model EOL on 7 July 2026.

The Sonnet 5.5 card dated 28 September 2026 lists 1 million tokens of context, 128,000 output tokens and a June 2026 cutoff. The EU profile is eu.anthropic.claude-sonnet-5-5. Model EOL is no sooner than 28 September 2027. Which Claude model to start with is in Claude Sonnet vs Opus.

The Llama 3.2 3B card lists launch on 25 September 2024, model EOL on 7 July 2026, 128,000 tokens of context, 4,000 output tokens and runtime ID meta.llama3-2-3b-instruct-v1:0. The EU profile is eu.meta.llama3-2-3b-instruct-v1:0. The 1B card shares that EOL and the profile eu.meta.llama3-2-1b-instruct-v1:0. Both cards mark the lifecycle as Legacy. Today, 15 October 2026, that EOL is more than three months past.

Trait	Claude Sonnet 5.5	Llama 3.2 3B	Llama 3.3 70B	Llama 4 Scout	Llama 4 Maverick
Role	production, agents	light Legacy model	dense 70B	long-context MoE	multimodal MoE
Launch	28 September 2026	25 September 2024	6 December 2024	5 April 2025	5 April 2025
Context / output	1M / 128K	128K / 4K	128K / 4K	10M / 8K	1M / 8K
Bedrock ID	eu.anthropic.claude-sonnet-5-5	eu.meta.llama3-2-3b-instruct-v1:0	us.meta.llama3-3-70b-instruct-v1:0	us.meta.llama4-scout-17b-instruct-v1:0	us.meta.llama4-maverick-17b-instruct-v1:0
EU on Bedrock	geo profile, Active	geo profile, EOL 7 July 2026	US geo only	US geo only	US geo only
Weights	closed	Community License	Community License	Community License	Community License

The Scout card lists 17 billion active parameters, 16 experts, 10 million tokens of context and 8,000 output tokens. The Maverick card lists 17 billion active, 128 experts and 1 million tokens of context. Both launched on 5 April 2025, both have EOL no sooner than 5 April 2026, and both carry model EOL on 30 March 2027. The 3.3 70B card lists launch on 6 December 2024, 128,000 tokens of context, 4,000 output tokens and only the US profile us.meta.llama3-3-70b-instruct-v1:0.

The Hugging Face Scout card lists 109 billion parameters in total, about 40 trillion training tokens and the Llama 4 Community License. The extra commercial term applies if you or your affiliates had more than 700 million monthly active users on the release date. The 3.3 card lists 70 billion parameters, 128k context, more than 15 trillion tokens and the same 700 million threshold. This is not Apache 2.0. The Apache 2.0 open-weight comparisons are Qwen vs Claude and Mistral vs Claude.

What does Llama vs Claude cost in the EU?

Llama vs Claude is uneven on the price page. Anthropic lists Sonnet 5.5 at 2 / 10 USD per million tokens, regional endpoints 10 percent higher, so 2.20 / 11 USD in the EU. The Llama 4 and Llama 3.3 accordions on the Bedrock pricing page rendered no dollar table in this run. We therefore set no on-demand rate for Scout, Maverick or 3.3.

Token prices on 15 October 2026, standard on-demand, USD per million tokens:

Route	Input / output	Cache read	Note
Claude Sonnet 5.5, Claude API global	2.00 / 10.00	0.10	5-minute write 2.50, 1-hour write 4.00
Claude Sonnet 5.5, Bedrock EU or Google Europe	2.20 / 11.00	0.11 (10 percent rule)	Batch 50 percent of the EU row
Claude Opus 5.5, Claude API global	4.00 / 20.00	0.20	regional endpoints plus 10 percent
Claude Haiku 5.5, prompts up to 100,000 tokens	0.10 / 0.50	0.01	above that 0.50 / 2.50
Llama 4 Scout / Maverick, Bedrock on-demand	not read	not read	accordion without a table
Llama 3.3 70B, Bedrock on-demand	not read	not read	accordion without a table
Llama 2 Chat 70B, US East / West	1.95 / 2.56	not read	only Meta dollar table this page showed
Custom Model Import, Frankfurt, 8 units (70B reference)	0.57152 USD / minute	n/a	8 times 0.07144, our sum

The pricing page sets Priority at a 75 percent premium and Flex and Batch at a 50 percent discount to Standard. Custom Model Import in Frankfurt costs 0.07144 USD per Custom Model Unit per minute, storage 1.95 USD per unit per month. The reference row names 8 units for Llama 3.1 70B 128K and 2 units for Llama 3.1 8B 128K. Eight units are 0.57152 USD per minute, about 34.29 USD per active hour, or about 24,690 USD across 30 days if the model never scales to zero. Storage for 8 units: 15.60 USD per month. The full Claude matrix is in Claude API pricing in the EU.

Foundry lists Llama-4-Maverick-17B-128E-Instruct-FP8 at 1 million input and output tokens and Llama-3.3-70B-Instruct at 128,000 input and 8,192 output. That catalogue page has no public dollar rate. Google MaaS for Maverick also points at a pricing page with no figure on the model card.

Does inference stay in the EU?

Claude Sonnet 5.5 stays in the EU if you call eu.anthropic.claude-sonnet-5-5. Llama 4 does not stay in the EU on Bedrock: Scout and Maverick have us.meta. profiles only. The compatibility page lists us-east-1, us-east-2, us-west-1 and us-west-2 for both. Llama 3.3 70B lists us-east-1, us-east-2 and us-west-2.

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 on one data centre. How Bedrock geo profiles work is in the Amazon Bedrock EU guide.

The Llama 3.2 EU profiles routed from Frankfurt, Ireland and Paris only into those three regions. The compatibility page marks the geo cells as Legacy with EOL 7 July 2026. A new statement of work should not treat 1B or 3B as the EU default.

Llama 4 Scout, Llama 4 Maverick, Llama 3.2 11B and Llama 3.2 90B use geofencing: access depends on the country of the AWS account and the source IP. Streaming via InvokeModelWithResponseStream or ConverseStream is not available for Llama 4 Instruct.

Google Cloud lists Claude Sonnet 5.5 in the Europe multi-region at 1,250 requests per minute, 12,500,000 uncached input TPM and 1,250,000 output TPM. Llama 4 Maverick MaaS sits only in us-east5, ML processing is United States multi-region, context is 524,288 tokens, output is 8,192 tokens, release 29 April 2025. At most three images per request. No EU region is listed.

Foundry Global Standard lists Llama-3.3-70B-Instruct and Llama-4-Maverick-17B-128E-Instruct-FP8 in France Central, Germany West Central, Italy North, Poland Central, Spain Central, Sweden Central and West Europe, among others. Global Standard may process prompts in any region where the model is deployed. The Data Zone Standard table for other Foundry models in Europe showed no Llama row in this run. Claude has no EU data zone on Foundry in our reading; that is in the Foundry guide and the GDPR provider comparison.

When do you take Llama, when Claude?

Take Claude Sonnet 5.5 on Bedrock EU when prompts must stay in member states and you need agents, computer use or 1 million tokens of context. Take Llama 4 only when Foundry Global, self-hosting or a US region on Bedrock fits your DPA. Do not treat the Llama 3.2 EU IDs as a new default: their model EOL is 7 July 2026.

1. Write residency into the statement of work: one member state, EU geography or global.
2. Score the task on a fixed eval set, not a demo chat.
3. Pinning: Claude only through eu.anthropic.claude-sonnet-5-5 or Google region="eu". Llama 4 on Bedrock only with a deliberate US region and a geofence check.
4. Put a cost cap on a LiteLLM gateway so apps cannot silently flip to global. or us-east-1.
5. Plan Llama 4 against model EOL on 30 March 2027. Name the successor before Legacy status blocks procurement.
6. If weights must sit in your data centre, self-host Llama 4 or 3.3. The Scout card names transformers 4.51.0 or newer. Read the 700 million MAU clause before you redistribute weights.

We recommend Bedrock EU for Claude as the default in European companies. Llama 4 wins on weights and on Scout’s 10 million token context, not on residency on AWS or Google. If you want both (open weights and EU processing), use your own GPUs or confirm Foundry Data Zone in the console, not the bare Bedrock ID.

How do you call both models on Bedrock?

Two Converse calls are enough to pin Claude on the EU profile and Llama 4 in Virginia. We leave Llama 4 in us-east-1 on purpose so the region break stays visible. Set AWS_BEARER_TOKEN_BEDROCK or IAM, not an Anthropic key. Streaming is off for Llama 4.

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)
# Llama 4 has no EU region. Call only if the DPA allows US processing.
scout = complete("us-east-1", "us.meta.llama4-scout-17b-instruct-v1:0", prompt)
print(claude["model"], scout["region"])

1. Create the client in eu-central-1 when inference should start in Frankfurt.
2. Pass Claude the eu. profile, not global.anthropic.claude-sonnet-5-5.
3. Pass Llama 4 the us. ID in a US region from the compatibility page.
4. Log modelId and region next to the request ID, or FinOps only sees a Bedrock account.
5. For Llama in the EU, change path: your own hardware or Foundry after a region check, not Bedrock.

Self-hosting follows the Hugging Face card: transformers 4.51.0 or newer, model ID meta-llama/Llama-4-Scout-17B-16E-Instruct. We did not read EU rental prices for the GPUs this needs, so we set none.

FAQ

What does Llama 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. Haiku 5.5 up to 100,000 tokens costs 5 USD. The Bedrock pricing page did not show an on-demand rate for Llama 4 or 3.3 in this run. Custom Model Import for the 70B reference in Frankfurt is about 34.29 USD per active hour.

Llama vs Claude: which model should a European company take first?

Claude Sonnet 5.5 through eu.anthropic.claude-sonnet-5-5 when personal prompts must stay in the EU. Llama 4 first only for self-hosting, Foundry after a region check, or a DPA that allows US Bedrock. The Llama 3.2 EU IDs are not a new default after 7 July 2026.

Does Llama 4 stay in the EU on Amazon Bedrock?

No. The model cards and the region list on 15 October 2026 name only US regions and the us. geo prefix for Scout and Maverick. There is no eu.meta.llama4. profile. For EU processing of Llama 4 use your own hardware or check Foundry, not Bedrock.

Self-host Llama 4 or run it on Foundry?

Self-host when the weights must sit in your data centre. The Scout card names the Community License, 109 billion parameters in total and transformers 4.51.0 or newer. Foundry Global Standard lists Maverick in several European regions but may process inference worldwide. The Data Zone table for Europe showed no Llama row in this run.

Claude on Google or Llama 4 on Google?

Claude Sonnet 5.5 has a Europe multi-region at 1,250 QPM and 1 million tokens of context. Llama 4 Maverick MaaS sits in us-east5 with US multi-region processing, 524,288 tokens of context and 8,192 output. The same frontier comparison on Google in the EU has no Llama 4.

Does Llama 3.3 70B have an EU price on Bedrock?

On 15 October 2026 the pricing page showed no dollar table for Llama 3.3 or Llama 4. The compatibility page lists 3.3 in US regions only. We leave the EU dollar rate blank until AWS publishes an EU row.

Sources

1. Anthropic: Pricing (15 October 2026)
2. AWS: Claude Sonnet 5.5 Model Card (15 October 2026)
3. AWS: Llama 3.2 3B Instruct Model Card (15 October 2026)
4. AWS: Llama 3.2 1B Instruct Model Card (15 October 2026)
5. AWS: Llama 4 Scout 17B Instruct Model Card (15 October 2026)
6. AWS: Llama 4 Maverick 17B Instruct Model Card (15 October 2026)
7. AWS: Llama 3.3 70B Instruct Model Card (15 October 2026)
8. AWS: Regional availability by models (15 October 2026)
9. AWS: Meta Llama model parameters (15 October 2026)
10. Amazon Bedrock Pricing (15 October 2026)
11. Hugging Face: Llama-4-Scout-17B-16E-Instruct (15 October 2026)
12. Hugging Face: Llama-3.3-70B-Instruct (15 October 2026)
13. Google Cloud: Llama 4 Maverick 17B-128E (15 October 2026)
14. Google Cloud: Claude Sonnet 5.5 (15 October 2026)
15. Microsoft Learn: Foundry models sold by Azure (15 October 2026)
16. Microsoft Learn: Foundry region availability (15 October 2026)