Voyage AI pricing

About

Voyage AI is a research-led embedding-model company, founded by Stanford AI researchers and acquired by MongoDB in early 2025, where it now operates as the embedding and reranking layer for MongoDB’s vector-search stack. Its product is a family of API-served models — general-purpose text embeddings (the voyage-4 series and older voyage-3 series), domain-specific embeddings (finance, law, code), contextualized chunk embeddings (voyage-context-3), multimodal embeddings (voyage-multimodal-3.5 / 3), and rerankers (rerank-2.5 / 2.5-lite) — used to power retrieval, RAG, and search applications.

The company sells these models directly through a standalone usage-based API at docs.voyageai.com, and also distributes them via cloud marketplaces (AWS, Azure) and data platforms (Snowflake) for VPC deployment. It competes with the embedding/reranker endpoints of OpenAI, Cohere, and Google, positioning on retrieval quality at a lower per-token price point — voyage-4 is marketed as outperforming major competitors while costing less.

Pricing is fully public and self-serve: there are no seats, plans, or platform fees. Customers pay per million tokens (or per pixel for multimodal), starting only after a generous free-token allowance, with throughput governed by usage-based rate-limit tiers rather than paid plans.

Pricing summary : per-1M-token rates by model with a 200M free tier

Voyage AI uses pure usage-based pricing: every model is billed per million tokens processed (multimodal additionally bills per billion pixels), with no subscription, seat, or platform fee. The model you call sets the rate, and a large free-token allowance precedes any charge.

• Text embeddings — billed on document/query tokens: voyage-4-lite $0.02/1M, voyage-4 $0.06/1M, voyage-4-large $0.12/1M, voyage-context-3 and voyage-code-3 $0.18/1M (each with 200M free tokens); voyage-finance-2, voyage-law-2, and voyage-code-2 $0.12/1M (50M free tokens each).
• Multimodal embeddings — voyage-multimodal-3.5 / voyage-multimodal-3 billed at $0.12/1M text tokens plus $0.60/1B pixels, with 200M text tokens and 150B pixels free; per-image cost is clamped between $0.00003 (50K-pixel floor) and $0.0012 (2M-pixel ceiling).
• Rerankers — rerank-2.5 $0.05/1M and rerank-2.5-lite $0.02/1M, billed on total processed tokens — (query tokens × #documents) + sum of all document tokens — with 200M free tokens each.
• Batch API — a 33% discount versus standard endpoints, with a 12-hour completion window. Files API storage is $0.05/GB/month.
• Throughput — governed by usage-based rate-limit tiers (Tier 1 default; ≥$100 paid → 2× limits; ≥$1,000 paid → 3×), not by a paid plan.

What makes this different: there is no “plan” to buy at all — the free-token allowance, per-model rate, and automatic rate-limit graduation replace the tiered-subscription structure most AI vendors use.

Pricing by product

Text embeddings (current models)

Multimodal embeddings

Rerankers

Batch & Files API

Sales motions across products: PLG / self-serve usage-based API for all standard embedding, multimodal, reranker, Batch, and Files models; sales-led for fine-tuned models and dedicated VPC instances.

Hidden costs : reranker double-billing, the two-meter multimodal bill, and rate-limit gating

The headline per-token rates are simple, but three things inflate a real Voyage bill: rerankers are billed on a multiplied token count, multimodal runs two meters at once, and throughput is capped until cumulative spend graduates your account.

Archetype 1 — a RAG app embedding + reranking at moderate scale. A team indexes 500M tokens/month of documents with voyage-4 and reranks 2,000 queries/day, each over 100 candidate documents of ~500 tokens.

The lesson: the reranker is the dominant line item, not the embeddings. Because reranker billing uses (query tokens × #documents) + sum of document tokens, a 100-document rerank of 500-token chunks bills ~50,000 tokens per request even though the query itself is tiny — so reranking 60,000 queries/month quietly out-costs embedding 500M tokens.

Archetype 2 — a multimodal search product over an image catalog. A team embeds 2M product images/month (avg 1M pixels each) plus 200M text tokens of captions on voyage-multimodal-3.5.

The lesson: for image-heavy workloads the pixel meter dominates and the text-token free allowance is almost irrelevant. Each image is clamped to a $0.0012 ceiling (2M pixels), so very high-resolution images don’t run away — but volume does, and the 150B-pixel free allowance is exhausted by ~150,000 one-megapixel images.

Want to estimate your own Voyage AI bill? Use the Voyage AI pricing calculator to model your monthly cost across embedding tokens, reranker requests, and multimodal pixels. For more on metering models like this, see our guide to choosing a usage-based pricing metric and our post on token-based vs outcome-based billing.

Pricing evolution : from a 50M-token alpha to a 200M-token free tier through the MongoDB acquisition

Voyage’s per-token rates have been remarkably stable since the alpha — the changes have been new model generations layered in at the same price points, a doubled-then-quadrupled free allowance, and an acquisition that left standalone pricing untouched.

Cadence

Tracked range: 2024-02–2026-06. Quarters not listed above were verified stable (0 price changes, 0 SKU additions).

Notable changes

• 2024-02 — Alpha pricing: voyage-2 at $0.10/1M with a 50M-token free allowance; no rerankers or multimodal models (docs.voyageai.com/docs/pricing, Wayback 2024-02-27).
• 2024-11 — voyage-3 series cut the flagship general model to $0.06/1M, quadrupled the free allowance to 200M tokens, and introduced multimodal and reranker product lines in one wave (Wayback 2024-11-26).
• 2025-02-24 — MongoDB acquired Voyage AI for approximately $220M; the standalone usage-based API, marketplace listings, and public per-token pricing were preserved as Phase 1 of a three-phase integration.
• 2025-04 — voyage-3-large and voyage-code-3 added at $0.18/1M post-acquisition, with no change to the free allowance (Wayback 2025-04-19).
• 2025-12 — voyage-3.5 refresh, contextualized voyage-context-3, multimodal-3.5 (video), and rerank-2.5 added; Batch API 33% discount and Files API storage published (Wayback 2025-12-27).
• 2026-06 — voyage-4 / 4-lite / 4-large supersede the voyage-3.5 generation at the same price points (docs.voyageai.com/docs/pricing, captured 2026-06-04).

The MongoDB acquisition in detail

The clearest pricing signal in Voyage’s history is a non-event: an acquisition that did not move prices. MongoDB paid roughly $220M in February 2025 and structured the integration in three phases. Phase 1 explicitly kept Voyage’s own website, AWS Marketplace, and Azure Marketplace channels live with unchanged usage-based pricing; the archived pricing page two months later (Wayback 2025-04-19) confirms identical per-token rates and the same 200M-token free allowance. Phase 2 embedded the models into MongoDB Atlas (auto-embedding for Vector Search, then native reranking), and Phase 3 targets multi-modal and instruction-tuned retrieval. For a standalone API buyer, the practical effect was additive distribution — Atlas became another way to consume the models — rather than a repricing or a forced migration, a contrast with acquisitions that fold a product behind a parent’s subscription.

What’s unique : no plans, a 200M-token free tier, and dual-meter multimodal billing

There is no plan to buy. Unlike most AI vendors, Voyage has no subscription, seat, or platform fee anywhere in its standard offering — you call a model, you pay that model’s per-token rate, and that’s the entire bill. The tiered-subscription scaffolding is replaced by a per-model rate plus an automatic, spend-driven rate-limit graduation. This is pure usage-based pricing in its cleanest form.

The free allowance is unusually generous and per-account. Every account gets 200M free tokens on current models — orders of magnitude larger than typical free credits — and the allowance persists after a card is added rather than expiring at signup. That makes Voyage genuinely free to evaluate at production-representative scale, a deliberate freemium acquisition lever.

Multimodal bills on two meters at once. Text-plus-image models charge $0.12/1M text tokens AND $0.60/1B pixels simultaneously, with per-image cost clamped to a $0.00003 floor and $0.0012 ceiling and each video frame billed as an image. This is one of the few embedding products that exposes a pixel-based meter alongside tokens, and it makes image-heavy workloads price very differently from text.

Throughput is earned, not purchased. Rate limits scale with cumulative spend — $100 billed unlocks 2x, $1,000 unlocks 3x — and tiers never downgrade. Prepaid credits accelerate the climb. Capacity is therefore a function of loyalty/volume rather than a line item, which removes a common enterprise upsell.

Same price across model generations. Each new flagship generation (voyage-3 → voyage-3.5 → voyage-4) has shipped at the same per-token price as the model it replaced, so customers get quality improvements without a price increase — a notable discipline through an acquisition.

Strengths & weaknesses

Billing UX : monthly post-paid usage with prepaid credit option and rate-limit tiers

• Monthly usage billing — usage is metered per token (and per pixel for multimodal) and billed monthly; customers can expect a credit-card charge around the 2nd of each month for the prior month’s usage.
• Free-token allowance — the first 200M tokens (current models; 50M for the voyage-*-2 domain models) are free for every account and persist even after a payment method is added.
• Prepaid usage credits — customers can pre-purchase a credit balance (with optional automatic recharge below a threshold) that is drawn down before metered billing; purchased credits expire after 1 year and are non-refundable.
• Usage-based rate-limit tiers — Tier 1 (payment method added) sets base TPM/RPM; ≥$100 billed/credited graduates to Tier 2 (2× limits) and ≥$1,000 to Tier 3 (3× limits); tiers never downgrade, and purchased credits accelerate progression.
• Organization & project rate limits — limits apply org-wide and can be sub-allocated per project by an Admin via the dashboard’s Rate Limits section.
• Zero-day data retention opt-out — Admins can toggle opt-out so Voyage does not store or train on submitted data.

Strategic wins : the decisions that made Voyage’s pricing work

1. A 200M-token free tier turns evaluation into adoption

By giving every account 200M free tokens — and letting that allowance persist after a card is added — Voyage lets teams embed a real corpus before paying a cent. This converts the riskiest part of an embedding migration (re-vectorizing everything) into a free trial, and it is a textbook freemium acquisition lever for a developer-first product. The free allowance is large enough that many side projects never reach a bill at all.

2. Holding price flat across model generations builds trust

voyage-3, voyage-3.5, and voyage-4 each shipped at the same per-token rate as the model they replaced. Customers got better retrieval quality without a price bump or a forced re-contract — a discipline that’s rare in AI, where new flagship models usually command a premium. This makes upgrading a no-brainer and removes the “should I wait for a price change?” hesitation. It’s a strong example of decoupling value delivery from price increases.

3. Spend-based rate-limit graduation replaces an enterprise upsell

Instead of selling “higher limits” as a plan, Voyage graduates throughput automatically as cumulative spend crosses $100 and $1,000. Growing customers get more capacity as a reward for usage rather than a negotiation, which keeps the self-serve motion intact deep into the adoption curve and defers sales involvement to genuinely custom (fine-tuned / VPC) needs.

4. Surviving acquisition with pricing intact protected the install base

When MongoDB acquired Voyage for ~$220M, the obvious risk for standalone API customers was a forced migration or repricing. Keeping the API, the marketplaces, and the public rates unchanged as Phase 1 retained those customers and added Atlas as a new distribution path rather than a replacement — a model for how to acquire an API company without churning its developers.

Areas to improve : where Voyage’s pricing creates friction

1. Reranker billing is hard to predict — publish a clearer estimator

Because rerankers bill on (query tokens × #documents) + sum of document tokens, the headline $0.05/1M rate badly understates real reranking cost; a 100-document rerank can bill ~50K tokens per request. Voyage already publishes a single “estimated price per request” figure, but an interactive estimator keyed on document count and chunk size would prevent the surprise our Hidden-costs archetype shows. Clearer up-front math here would reduce bill shock.

2. Make the dual-meter multimodal bill legible before the invoice

The token-plus-pixel multimodal model is powerful but opaque: teams must reason about two meters and a per-image clamp to forecast spend. A worked “cost per 1,000 images at resolution X” widget on the pricing page (beyond the static examples) would let image-search builders budget confidently and avoid the pixel-meter dominating unexpectedly.

3. Offer a published throughput/SLA option without a sales call

Rate limits graduate with spend, but a new account starts at Tier 1 with no self-serve way to buy higher limits or a committed SLA short of engaging sales for a dedicated instance. A published, self-serve “guaranteed throughput” add-on would serve mid-market teams who need predictability but aren’t ready for a VPC contract — closing a gap between self-serve and sales-led motions.

4. Reconsider 1-year, non-refundable prepaid credit expiry

Prepaid credits accelerate rate-limit progression but expire after one year and are non-refundable. For lumpy or seasonal workloads this is a quiet penalty. Rolling expiry or partial refundability would make prepayment a friendlier commitment device rather than a use-it-or-lose-it risk.

Key takeaways

1. A large, persistent free tier can replace a trial entirely. Voyage’s 200M free tokens per account — surviving even after a card is added — lets teams embed a production corpus for free, turning the costliest migration step into adoption fuel. Other API companies under-size free tiers and lose evaluators at the re-vectorization wall.
2. Holding price flat across generations is a retention strategy. Shipping voyage-3 → 3.5 → 4 at identical per-token rates makes every upgrade frictionless and removes price-change hesitation. Quality, not price, becomes the upgrade trigger.
3. Throughput can be a loyalty reward instead of an upsell. Graduating rate limits by cumulative spend ($100 → 2x, $1,000 → 3x) keeps self-serve intact and defers sales to genuinely custom needs, avoiding a per-plan “limits” tax.
4. Watch for meters hidden inside a “simple” per-token price. Reranker token multiplication and multimodal pixel billing mean the headline rate isn’t the bill — pricing teams should stress-test the actual metering formula against realistic workloads.
5. An acquisition is a pricing decision. Keeping standalone pricing, channels, and free tiers intact post-acquisition retained Voyage’s developer base; folding it behind a parent subscription would have churned it. How you reprice (or don’t) after M&A is a first-class strategy choice.

UBP implications

1. Pure usage with no seats scales cleanly from hobbyist to enterprise. Voyage shows a metered API can run the full customer journey on one price list — a generous free band, per-unit rates, and spend-based capacity — without ever introducing plans. This is the model usage-based AI infrastructure increasingly converges on.
2. Multi-meter billing is becoming standard for multimodal AI. Charging tokens and pixels simultaneously (with per-unit clamps) previews how usage-based pricing handles heterogeneous inputs; UBP designers should plan for compound meters rather than a single value metric as products go multimodal.
3. Free-tier generosity is a defensible moat in commodity-adjacent markets. When embedding quality and price are close across vendors, a 200M-token persistent free allowance becomes a real acquisition advantage — a reminder that in UBP the shape of the free band can matter as much as the per-unit rate.

Sources

• Voyage AI pricing — primary per-token, multimodal, reranker, Batch, and Files API rates (accessed 2026-06-04)
• Voyage AI rate limits — TPM/RPM base limits and the $100/$1,000 usage-tier progression (accessed 2026-06-04)
• Voyage AI FAQ / docs — free-tier statement, monthly billing cycle, and prepaid-credit terms (accessed 2026-06-04)
• Voyage AI homepage — model lineup and positioning (accessed 2026-06-04)
• Voyage AI blog — voyage-3.5 — independent second-source confirmation of the $0.06 / $0.02 per-1M-token rates that voyage-4 / voyage-4-lite inherit (accessed 2026-06-04)
• Voyage AI docs — Text Embeddings — independent confirmation of the voyage-4 / 4-lite / 4-large model lineup and recommendations (accessed 2026-06-04)
• Historical pricing verified against docs.voyageai.com/docs/pricing snapshots in the Internet Archive Wayback Machine (2024-02-27 through 2025-12-27); see the Blueprint corpus for methodology.

Bottom line

Voyage AI is one of the cleanest examples of pure usage-based pricing in AI infrastructure: no seats, no plans, a 200M-token free allowance per account, and per-model rates that have held flat across three model generations — and even through a ~$220M MongoDB acquisition that left the standalone API untouched. The catch is in the meters, not the rates: rerankers bill on multiplied token counts and multimodal models run a pixel meter alongside tokens, so the real bill rewards teams that model their actual workload before committing. Compare its approach with peers in the Blueprint corpus.

Pricing timeline : Major events on a vertical axis

Each milestone below corresponds to a public pricing change, product launch, or material adjustment. Major events use a filled marker; minor adjustments use a faded one.

Current pricing snapshot (voyage-4 series)

Jun 2026

Usage-based per-1M-token pricing per model: voyage-4-lite $0.02, voyage-4 $0.06, voyage-4-large $0.12 (each replacing the corresponding voyage-3.5 model at the same price), voyage-context-3 / voyage-code-3 $0.18; domain models (finance/law/code-2) $0.12; multimodal-3.5/3 $0.12/1M tokens + $0.60/1B pixels; rerank-2.5 $0.05, rerank-2.5-lite $0.02. 200M free tokens (50M for voyage-*-2), 33% Batch API discount, Files API $0.05/GB/mo, and ≥$100/≥$1,000 usage-based rate-limit tiers.

voyage-3.5 refresh, voyage-context-3, multimodal-3.5, rerank-2.5

Dec 2025

By Wayback 2025-12-27: voyage-3.5 and voyage-3.5-lite replaced voyage-3/3-lite at the same $0.06 / $0.02 rates; contextualized voyage-context-3 added at $0.18/1M; multimodal-3.5 added (each video frame billed as an image); rerankers refreshed to rerank-2.5 ($0.05) and rerank-2.5-lite ($0.02). Batch API (33% off) and Files API ($0.05/GB/mo) storage formalised.

voyage-3-large and voyage-code-3 added at $0.18/1M

Apr 2025

Post-acquisition (Wayback 2025-04-19): higher-quality voyage-3-large and code-specialised voyage-code-3 added at $0.18/1M, both keeping the 200M free-token allowance. Existing voyage-3 ($0.06) and voyage-3-lite ($0.02) rates held flat.

MongoDB acquires Voyage AI (~$220M) — pricing unchanged

Feb 2025

MongoDB announced the acquisition of Voyage AI for ~$220M on 2025-02-24. Phase 1 of the integration explicitly preserved the standalone API: voyageai.com, AWS Marketplace, and Azure Marketplace access continued, and the Wayback 2025-04-19 pricing page shows public per-token rates unchanged by the deal. Models additionally began appearing inside MongoDB Atlas.

voyage-3 series, multimodal, and rerankers launch

Nov 2024

By Wayback 2024-11-26 the lineup expanded sharply: voyage-3 at $0.06/1M and voyage-3-lite at $0.02/1M with the free allowance raised to 200M tokens; domain models voyage-finance-2 / voyage-law-2 / voyage-multilingual-2 / voyage-code-2 at $0.12/1M (50M free); first multimodal model voyage-multimodal-3 at $0.12/1M text tokens + $0.60/1B pixels (200M tokens + 150B pixels free); and first rerankers rerank-2 ($0.05/1M) and rerank-2-lite ($0.02/1M), each 200M free.

Alpha usage-based pricing (voyage-2 series)

Feb 2024

Earliest archived pricing (Wayback 2024-02-27): per-token billing with voyage-2 at $0.10/1M, voyage-large-2 and voyage-code-2 at $0.12/1M, each with 50M free tokens; older voyage-01/02 models at $0.10/1M with no free tokens. No rerankers and no multimodal models yet; fine-tuned models were already sales-led.