Mistral Large
Mistral AI
★ #21 in Conversational AIProduct status: Status not recorded
No lifecycle review is recorded for this listing. Its inclusion in the directory does not confirm current availability; check the vendor before relying on it.
No status check date recorded
Mistral AI's versioned Large model line, with API access and release-specific weight availability, licenses and data controls. No lifecycle review is recorded for this listing. Its inclusion in the directory does not confirm current availability; check the vendor before relying on it.
Product preview
Decision notes
Official-document research
Based on the cited documentation. We did not use the product, purchase a plan, or run quality, speed or reliability tests in this review.
Consider it for
- Developers able to pin models and inspect outputs
- Teams checking a specific deployment's data terms
Choose another approach if
- Confidential data assuming preview opt-out/ZDR
- Assuming all Large releases share Apache or EU residency
Pricing and conditions
Documented terms
The USD Standard/global API table lists Large 4 sale rates of $0.68 input, $0.07 cached input and $2.09 output per million tokens, against struck-out $1.36/$0.14/$4.18; the promotion's end is unconfirmed. Large 3 is separately $0.50/$0.05/$1.50 per million. Taxes are extra. Regional inference is billed at 1.1× standard list pricing for input/output and cache reads/writes; promotion stacking is unconfirmed. Batch/Priority and provider contracts differ. No consumer subscription or seat quota is included. Self-hosted Large 3 incurs infrastructure costs; future Large 4 weights do not establish a current free deployment.
Pricing source read:
Pricing or terms sourceCode, model and service licensing
Large 3's linked card applies Apache-2.0 to that checkpoint; its license/notice, modification and patent conditions do not license the hosted API or unpublished Large 4 weights. Commercial terms retain relative data/output rights and assign only Mistral's rights, excluding weights and third-party references and guaranteeing no uniqueness. Processing, feedback and training permissions differ: Labs/Preview terms permit training and say other opt-out/ZDR choices do not apply, material for Large 4's current preview. ZDR documentation has narrower endpoint/account eligibility and excludes stateful services; obtain written confirmation where its Labs wording and broader Preview contract need reconciliation. Regional inference is separate from retention and control-plane residency. Input, output and third-party rights still require review. Labs/Preview access is exclusively for commercial customers' professional testing, not assumed production use.
Basis and limits of the recorded score
The existing 4.5 score is retained; its original scoring evidence is unavailable. This documentation review did not measure performance or compare products and did not re-score it.
Overview
Mistral Large names a changing model line. The current Large 4 API is a public preview, while Large 3 has a separate downloadable Apache-2.0 checkpoint. As of this review, the official Large 4 weight page still schedules an upcoming October 31, 2026 release; the API preview is not proof that those files or their license are already available. Pin the actual model ID instead of treating a latest alias as a fixed release.
Capabilities, cost and data processing follow that choice. Large 4 documents image/text tasks, structured outputs and function calling, but preview behavior may change. Regional inference is an optional eligible-endpoint setting, with an upcharge and feature limits; it does not place all account, billing or operational metadata in that region. The former blanket EU-sovereignty and competitor-parity claims are not supported here.
Sources
The factual claims in this review were read from the pages below. Each date is when an editor last opened the source, not when the product changed.
- Vendor documentation Checked 2026-10-15
- Vendor documentation Checked 2026-10-15
- Vendor documentation Checked 2026-10-15
- Vendor documentation Checked 2026-10-15
- Third-party report Checked 2026-10-15
- Vendor pricing page Checked 2026-10-15
- Vendor documentation Checked 2026-10-15
- Vendor documentation Checked 2026-10-15
- Vendor documentation Checked 2026-10-15
- Vendor documentation Checked 2026-10-15
Pros / Cons
Pros
- Published model-specific capabilities and API units
- Large 3 offers a separate Apache-licensed weight path
- Optional regional inference controls
Cons
- Large 4 preview can change and permits training under terms
- Large 4 weight availability is still scheduled
- Region and retention settings are not universal
Who it's for
Evaluating a pinned multimodal API
Comparing Large 3 self-hosting with hosted releases
Bottom line
Select generation, infrastructure and contract together; Large 3's weight license does not establish Large 4's preview rights.
Key Features
Tags
Frequently asked questions
Is Mistral Large currently available?
No lifecycle review is recorded for this listing. Its inclusion in the directory does not confirm current availability; check the vendor before relying on it.
What is Mistral Large?
Mistral Large is a Conversational AI tool developed by Mistral AI. Mistral AI's versioned Large model line, with API access and release-specific weight availability, licenses and data controls.
What costs and terms should I check for Mistral Large?
The USD Standard/global API table lists Large 4 sale rates of $0.68 input, $0.07 cached input and $2.09 output per million tokens, against struck-out $1.36/$0.14/$4.18; the promotion's end is unconfirmed. Large 3 is separately $0.50/$0.05/$1.50 per million. Taxes are extra. Regional inference is billed at 1.1× standard list pricing for input/output and cache reads/writes; promotion stacking is unconfirmed. Batch/Priority and provider contracts differ. No consumer subscription or seat quota is included. Self-hosted Large 3 incurs infrastructure costs; future Large 4 weights do not establish a current free deployment.
Is Mistral Large open source?
Large 3's linked card applies Apache-2.0 to that checkpoint; its license/notice, modification and patent conditions do not license the hosted API or unpublished Large 4 weights. Commercial terms retain relative data/output rights and assign only Mistral's rights, excluding weights and third-party references and guaranteeing no uniqueness. Processing, feedback and training permissions differ: Labs/Preview terms permit training and say other opt-out/ZDR choices do not apply, material for Large 4's current preview. ZDR documentation has narrower endpoint/account eligibility and excludes stateful services; obtain written confirmation where its Labs wording and broader Preview contract need reconciliation. Regional inference is separate from retention and control-plane residency. Input, output and third-party rights still require review. Labs/Preview access is exclusively for commercial customers' professional testing, not assumed production use.
What are the best alternatives to Mistral Large?
Notable Conversational AI alternatives include ChatGPT, Claude, Gemini. Browse all 67 tools in our Conversational AI category for a full overview.
Was this listing useful?
Basic Info
- Pricing
- Pay-as-you-go
The USD Standard/global API table lists Large 4 sale rates of $0.68 input, $0.07 cached input and $2.09 output per million tokens, against struck-out $1.36/$0.14/$4.18; the promotion's end is unconfirmed. Large 3 is separately $0.50/$0.05/$1.50 per million. Taxes are extra. Regional inference is billed at 1.1× standard list pricing for input/output and cache reads/writes; promotion stacking is unconfirmed. Batch/Priority and provider contracts differ. No consumer subscription or seat quota is included. Self-hosted Large 3 incurs infrastructure costs; future Large 4 weights do not establish a current free deployment.
- Editorial score
- ★ 4.5 / 5 How we score
- Platforms
- APIWeightsSelf-hosted
- Last verified
- Not recorded
