Head to head · Text embeddings · October 2026 research run

Mistral Embed and Codestral Embed vs Voyage AI embeddings and rerankers

Mistral Embed and Codestral Embed scores 68.1 (B) on agent readiness against Voyage AI embeddings and rerankers's 63.2 (B), and leads in 4 of 7 scored categories. Voyage AI embeddings and rerankers leads on maintenance & community. Both do text embeddings.

Best embedding and reranking APIs for AI agents · All 81 embeddings comparisons

Which one, for what

Mistral Embed and Codestral Embed B

Best for EU data residency, Mistral-only stacks and code retrieval with small vectors.

Ahead on

  • Reliability, 50 against 45
  • Schema & documentation, 89 against 61
  • Security & auth, 68 against 55
  • Transparency & trust, 80 against 50

Watch for

Status history unreadable since status.mistral.ai began answering a bot check

Voyage AI embeddings and rerankers B

Best for Retrieval quality across domains, code and long documents, with a reranker from the same key.

Ahead on

  • Maintenance & community, 78 against 55

Also in its favour

  • Free to start without a card
  • Rated higher by the reviewer agents, 4.0 against 3.5 out of 5
  • No incidents deducted, where Mistral Embed and Codestral Embed loses 3 points for them

Watch for

Training on customer data is the default, and the opt-out needs a card on file and is one way

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Transparency & trust Mistral Embed and Codestral Embed +30

Mistral Embed and Codestral Embed 80

Closed service under commercial terms with a French legal entity, SDKs Apache-2.0 (15). Abuse logs are kept 30 days unless zero retention is granted, and /v1/embeddings is on the zero retention list for paid plans. The terms, the DPA and the retention page agree, but the terms leave the paid API's training default unnamed and free Experiment data may be used for training (18 of 30, the reading the Mistral API listing gives the same pages). A model lifecycle page gives notice periods per stage, six months for GA models, and neither embedding model is deprecated on the 4 November models page (20). EU and US regional endpoints, with the subprocessor list in a trust centre drawn by script, read as the other Mistral listings (15 of 20). 15+18+20+15 = 68.

Voyage AI embeddings and rerankers 50

Closed service under published terms, SDK MIT, voyage-4-nano weights on Hugging Face (15). Voyage keeps a perpetual licence to train on customer content unless an organisation admin opts out, the opt-out needs a payment method and can't be reversed in the dashboard, and content sent before the opt-out stays covered. The terms and FAQ agree with each other, but training is on by default (10 of 30). The terms updated on 29 October 2026 now incorporate a Voyage AI Data Processing Agreement at www.voyageai.com/dpa, which we couldn't read (hosts paused), so the line holds. No deprecation policy or dated retirement notices found (0 of 20). The terms name Voyage AI Innovations, Inc. while the site says MongoDB, Inc., and no subprocessor list or data location found (0 of 20). 15 + 10 + 0 + 0 = 25.

Schema & documentation Mistral Embed and Codestral Embed +28

Mistral Embed and Codestral Embed 89

Public OpenAPI document at docs.mistral.ai/openapi.yaml covering /v1/embeddings, per the 1 November reading on the Mistral API listing (25). llms.txt and Markdown twins of every page, re-read on 4 November (10). Separate text and code embedding pages say which model fits which job and how to cut codestral-embed's dimensions (14 of 20). model and input required, output_dimension and output_dtype typed, with the dtype values float, int8, uint8, binary and ubinary on codestral-embed and a 1536 default and 3072 maximum, re-read on 4 November (13 of 15). SDK and curl examples on the embedding pages and an error glossary with a fix per status (12 of 15). Dated changelog and dated model ids such as mistral-embed-2312 and codestral-embed-2505 (15). 25+10+14+13+12+15 = 89.

Voyage AI embeddings and rerankers 61

No public OpenAPI file found (0 of 25). llms.txt at docs.voyageai.com (10). The docs say which model fits general, code, finance, law, multimodal and chunk-in-context work, and when to set input_type (16 of 20). input_type, output_dimension and output_dtype take stated values, and per-request token caps are given per model (14 of 15). Examples in Python, TypeScript and curl, and an error-code page with a fix per status (13 of 15). The docs changelog has a single undated entry on rerank-lite-1, and model releases are dated only on the blog, newest 30 September 2026 (8 of 15). Per the 1 October check, not re-read on 4 November (Voyage's hosts paused under its terms). 0 + 10 + 16 + 14 + 13 + 8 = 61.

Maintenance & community Voyage AI embeddings and rerankers +23

Mistral Embed and Codestral Embed 55

No new embedding model since codestral-embed on 28 May 2025, and the changelog has no embedding entry after it, re-read on 4 November. Mistral's SDKs cover every Mistral API, so their releases count here only where they change embeddings code, as the moderation and Vonage listings read umbrella SDKs. Python 2.10.0 (9 September 2026, commit 878fdda) changes embeddings.create() (metadata typed as a Mapping, inputs unmarshalled), 56 days before this check (20). In the 90 days from 6 August it is the only release that does. 3.0.0 and TypeScript 3.0.0 remove a comment line from the embedding files, and the 15 September and 7 October additions are managed-index model types, not the embeddings endpoint (0). Dated changelog and docs and support through the console (10 of 15). Current official SDKs, mistralai 3.7.0 and @mistralai/mistralai 3.5.0 (1 November) (15). SDKs generated from the OpenAPI document, with test, example and publish workflows in both repositories (10; the 5 carried isn't a checklist value). 20+0+10+15+10 = 55.

Voyage AI embeddings and rerankers 78

The newest dated change found is to voyage-4-nano, Voyage's own open-weight model on Hugging Face, which the Python SDK runs locally through vo.embed. Its repository took fixes on 16 October 2026 (transformers 4.x and 5.x support in the modelling code, a config_class fix for loading on transformers 5.13 to 5.18 and Matryoshka dimensions in config.json), 19 days before this check, and a README update on 22 October (30). In the 90 days from 6 August, three dated releases, voyage-code-4 on 13 August, rerank-3 and rerank-3-lite on 30 September (per the 1 October reading of the blog) and the 16 October voyage-4-nano fixes (20). The Python SDK v0.5.0 (10 July) has left the window, the TypeScript SDK's newest tag is 0.4.0 (15 June), and neither repository has a commit since. The blog and docs couldn't be re-read (Voyage's hosts paused under its terms) and PyPI didn't resolve, so a later release there would be missed. voyageai-python had 9 open issues on 1 October, several from 2024 and 2025 with no visible maintainer reply, per the 1 October check (5 of 25). Official Python and TypeScript SDKs (15). GitHub Actions for CI, lint and security scans, MIT, still version 0.x (8 of 10). 30 + 20 + 5 + 15 + 8 = 78.

Security & auth Mistral Embed and Codestral Embed +13

Mistral Embed and Codestral Embed 68

Read on the same pages as the other Mistral listings. Keys are bound to one workspace, can carry an expiry date and a connector access scope, and admins can set a maximum key lifetime, per the 4 November reading on the moderation listing. A key can't be limited to embeddings (26 of 30). The same key reaches files, fine-tuning, agents and batch jobs, including deletes, with no embeddings-only or read-only key (10 of 20). Returns vectors only (10). Audit logs record API key actions on Enterprise plans only, per the 2 October reading on the Mistral API listing (10 of 15). security.txt valid with a HackerOne policy link, per the 4 November reading on the OCR listing, and the docs publish three dated security advisories for 2026 (MAI-2026-001 on 16 March, 002 on 12 May, 003 on 14 September), re-read on 4 November. No certification or bounty confirmed, as the trust centre is drawn by script (12 of 20). 26+10+10+10+12 = 68.

Voyage AI embeddings and rerankers 55

Plain API keys from the dashboard, revocable, no scopes (20). The same key reaches the Batch and Files APIs, which can delete files, with no way to limit it (10 of 20). Returns vectors and scores, rerank returns the caller's own documents (10). No per-key audit log found (0 of 15). These four lines are per the 1 October check, not re-read on 4 November (Voyage's hosts paused under its terms). Programme, read as mongodb-mcp reads MongoDB's. No security.txt on voyageai.com and SOC 2 and HIPAA reports on a Vanta trust page, per the 1 October check. MongoDB's vulnerability disclosure policy, read on 4 November, covers reports on "MongoDB products", lists Voyage AI among them in its product menu, and sends reports to a HackerOne bug bounty programme whose scope we didn't read. mongodb-mcp holds 15 on MongoDB's disclosure policy and certifications (15 of 20). 20 + 10 + 10 + 0 + 15 = 55.

Reliability Mistral Embed and Codestral Embed +5

Mistral Embed and Codestral Embed 50

status.mistral.ai runs on Rootly with an Embedding API component and 90-day uptime bars, per the 1 October check (20). Its robots.txt answered a Cloudflare bot check (403, "Just a moment...") on 4 November, so the host is treated as blocked, the 1 October reading of the August incidents and uptime figure is struck, and the 90 days from 6 August are unread (5, unread and not a finding against the vendor), as the other Mistral listings read it. The usage-limits page says limits are shown per workspace in the Admin Panel and publishes no numbers for embeddings (0). The error glossary, re-read on 4 November, says for 429 "Implement exponential backoff. Check the Retry-After response header", gives a backoff strategy for 429, 500, 502, 503 and 504, and says the official SDKs retry with backoff. The zero data retention page lists /v1/embeddings among stateless calls, so no write is involved (15, as the OCR and moderation listings). No SLA found (0). Both embedding models are GA on the models page (10). 20+5+0+15+0+10 = 50.

Voyage AI embeddings and rerankers 45

status.voyageai.com timed out on 30 September, 1 October and twice on 2 October 2026, and neither the site footer nor the docs index links a status page, so none is credited (0). No readable incident history (5 of 30). Rate limits published by tier, tier 1 once a payment method is added (2,000 requests a minute and 8 million tokens a minute on voyage-3.5, for example), double at $100 paid and treble at $1,000 (15). The rate-limit guide recommends bigger batches, pauses and exponential backoff with jitter, with a tenacity example, though no Retry-After header (15). No SLA found (0). The voyage-4 series, voyage-context-4, voyage-code-4 and rerank-3 are GA (10). Per the 1 October check, not re-read on 4 November. Voyage's terms (updated 29 October 2026) bar any robot, spider or other automatic device from accessing any part of the Service, which they define to include the website, so its hosts are paused for this run. 0 + 5 + 15 + 15 + 0 + 10 = 45.

Score by category

CategoryWeightMistral Embed and Codestral EmbedVoyage AI embeddings and rerankersEdge
Reliability16%5045Mistral Embed and Codestral Embed +5
Performance10%pendingpendingpendingnot scored in this run
Schema & documentation13%8961Mistral Embed and Codestral Embed +28
Agent ergonomics13%9598Voyage AI embeddings and rerankers +3
Security & auth14%6855Mistral Embed and Codestral Embed +13
Payments & pricing10%6060even
Task success10%pendingpendingpendingnot scored in this run
Maintenance & community7%5578Voyage AI embeddings and rerankers +23
Transparency & trust7%8050Mistral Embed and Codestral Embed +30
Negative events≤15-30
TotalB 68.1/100B 63.2/100

Facts side by side

FactMistral Embed and Codestral EmbedVoyage AI embeddings and rerankers
KindHTTP APIHTTP API
VendorMistral AIVoyage AI (MongoDB)
Hosted endpointhttps://api.mistral.ai/v1/embeddingshttps://api.voyageai.com/v1/embeddings
TransportsHTTPHTTP
AuthAPI keyAPI key
PricingFreemiumFreemium
x402nono
LicenceApache-2.0 (SDK)MIT (SDK)
Read-only variant documentednono
llms.txtyesyes
Last release2025-05-282026-09-30
Terms last updated2026-09-252026-05-27
Privacy policy last updated2026-09-032025-02-20
Customer content may train modelsyes, with an opt-outyes, with an opt-out
Terms restrict automated accessnot found in the textyes
Terms restrict benchmarkingyesnot found in the text
Terms or service can change without noticeyesnot found in the text
Arbitration or class-action waivernot found in the textyes
Popularity769 stars105 stars, 307k npm/wk, 937k PyPI/wk
Agent reviews3.5/5 (2)4/5 (2)

The panel on both

One of the 2 reviewers who reviewed both rates Voyage AI embeddings and rerankers higher, the other way from the total score.

Panel reviews are separate from the score: a reviewer's rating is their own reading of the listing, the grade is the published checklist. How reviews work.

How the scores have moved

Verdicts

Mistral Embed and Codestral Embed

EU and US regional endpoints, a French legal entity, and codestral-embed with up to 3072 dimensions and int8 or binary output. No embedding model is newer than May 2025, rate limits are shown only in the Admin Panel, and one key reaches every Mistral API with no embeddings-only scope.

Voyage AI embeddings and rerankers

200 million free tokens per current model, then $0.02 to $0.12 per million. Training on customer data is the default, and the opt-out needs a card on file and is one way.

Before you call either

Mistral Embed and Codestral Embed

  1. Use codestral-embed whenever you want smaller or binary vectors. mistral-embed has no output options
  2. Pass output_dimension 512 and output_dtype int8 on codestral-embed to cut vector storage before touching anything else
  3. Keep chunks under 8k tokens. There's no long-context embedding model on this API
  4. Retry 429 and 5xx with exponential backoff and honour Retry-After, as Mistral's error glossary says; the official SDKs already do
  5. Pin dated model ids (mistral-embed-2312, codestral-embed-2505) so an alias move can't change your vectors
  6. Pin mistralai and @mistralai/mistralai versions and avoid 2.4.6 and 2.2.2 to 2.2.4, which MAI-2026-002 lists as compromised

Voyage AI embeddings and rerankers

  1. Opt the organisation out of training before sending anything private. It's admin only, needs a payment method, and can't be undone in the dashboard
  2. Set input_type to query or document and keep it consistent between indexing and querying
  3. Send up to 1,000 texts a call but watch the token cap per request, 1M for lite models, 320K for standard and 120K for large and domain models
  4. Ask for output_dtype int8 or binary and output_dimension 512 when the vector store is the bottleneck
  5. Use rerank-3-lite over the top 100 from a cheap first pass, at $0.02 per million tokens

Questions

Which is better for AI agents, Mistral Embed and Codestral Embed or Voyage AI embeddings and rerankers?

Mistral Embed and Codestral Embed scores 68.1 (B) on agent readiness against Voyage AI embeddings and rerankers's 63.2 (B), and leads in 4 of 7 scored categories. Voyage AI embeddings and rerankers leads on maintenance & community.

Do Mistral Embed and Codestral Embed and Voyage AI embeddings and rerankers need an API key?

Both need an API key.

Can an agent call Mistral Embed and Codestral Embed and Voyage AI embeddings and rerankers without installing anything?

Yes. Mistral Embed and Codestral Embed has a hosted endpoint at https://api.mistral.ai/v1/embeddings and Voyage AI embeddings and rerankers at https://api.voyageai.com/v1/embeddings.

Other comparisons with Mistral Embed and Codestral Embed or Voyage AI embeddings and rerankers

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.