Embeddings
The retrieval layer. Input-only pricing, and the cheapest model is rarely the right one for code RAG.
Embedding models bill per million input tokens only, with no output meter. Cohere Embed 5 Pro and Fast are available through both Cohere Direct and Microsoft Foundry: Global text rates are $0.12/M and $0.08/M, while Data Zone rates are $0.132/M and $0.088/M. Image input has a separate rate and is outside the text-token comparison. Alibaba's Qwen3.7 text embedding adds a 128K-token multilingual option at $0.07/M, while OpenAI's small model remains the budget pick.
Pricing
per 1M tokens · USD / CHF| Model | Tier / Host | Context | Input | Confidence |
|---|---|---|---|---|
Cohere Embed 5 Pro High-quality multimodal retrieval model with a 128K-token context, 100+ languages, and 256–2048 output dimensions. Text input is $0.12/M; image input is $0.40/M and is outside this text-token comparison. Shares an embedding space with Embed 5 Fast. | Direct API Cohere API | 128K | $0.12CHF 0.097 | official Cohere's official Embed 5 launch post states that the model is available through Cohere API and Microsoft Foundry at $0.12 per 1M text tokens and $0.40 per 1M image tokens; Cohere's September 30 release notes identify the API model as embed-v5.0-pro. Captured 2026-10-09. This Direct row records the Cohere API text rate; Foundry rates from Azure's October 1 meters are listed separately below. Image input remains outside this text-token comparison. |
Cohere Embed 5 Fast Lower-latency, high-throughput multimodal retrieval model with a 128K-token context, 100+ languages, and 256–2048 output dimensions. Text input is $0.08/M; image input is $0.40/M and is outside this text-token comparison. Shares an embedding space with Embed 5 Pro. | Direct API Cohere API | 128K | $0.08CHF 0.064 | official Cohere's official Embed 5 launch post states that the model is available through Cohere API and Microsoft Foundry at $0.08 per 1M text tokens and $0.40 per 1M image tokens; Cohere's September 30 release notes identify the API model as embed-v5.0-fast. Captured 2026-10-09. This Direct row records the Cohere API text rate; Foundry rates from Azure's October 1 meters are listed separately below. Image input remains outside this text-token comparison. |
Cohere Embed 5 Pro Microsoft Foundry Direct from Azure Global Standard meter. The text-token rate matches Cohere's Direct API; image input is billed separately and is outside this comparison. | Foundry ·Global Direct from Azure | 128K | $0.12CHF 0.097 | official Azure Retail Prices API, serviceName 'Foundry Models', productName 'Cohere Models', captured 2026-10-09: effective 2026-10-01 meter 'Embed V5 Pro Txt Glbl' is $0.00012 per 1K tokens ($0.12/M). The Microsoft Foundry catalog identifies Cohere-Embed-V5-Pro as a Direct from Azure 128K multimodal embedding model. Its separate image meter is not represented in this text-token row. |
Cohere Embed 5 Pro Microsoft Foundry Direct from Azure Data Zone meter. The text-token rate is 1.1x Global; image input is billed separately and is outside this comparison. | Foundry ·Data Zone Direct from Azure | 128K | $0.132CHF 0.106 | official Azure Retail Prices API, serviceName 'Foundry Models', productName 'Cohere Models', captured 2026-10-09: effective 2026-10-01 meter 'Embed V5 Pro Txt DZ' is $0.000132 per 1K tokens ($0.132/M). The Global meter is $0.12/M. The Microsoft Foundry catalog identifies Cohere-Embed-V5-Pro as a Direct from Azure 128K multimodal embedding model. Its separate image meter is not represented in this text-token row. |
Cohere Embed 5 Fast Microsoft Foundry Direct from Azure Global Standard meter. The text-token rate matches Cohere's Direct API; image input is billed separately and is outside this comparison. | Foundry ·Global Direct from Azure | 128K | $0.08CHF 0.064 | official Azure Retail Prices API, serviceName 'Foundry Models', productName 'Cohere Models', captured 2026-10-09: effective 2026-10-01 meter 'Embed V5 Fast Txt Glbl' is $0.00008 per 1K tokens ($0.08/M). The Microsoft Foundry catalog identifies Cohere-Embed-V5-Fast as a Direct from Azure 128K multimodal embedding model. Its separate image meter is not represented in this text-token row. |
Cohere Embed 5 Fast Microsoft Foundry Direct from Azure Data Zone meter. The text-token rate is 1.1x Global; image input is billed separately and is outside this comparison. | Foundry ·Data Zone Direct from Azure | 128K | $0.088CHF 0.071 | official Azure Retail Prices API, serviceName 'Foundry Models', productName 'Cohere Models', captured 2026-10-09: effective 2026-10-01 meter 'Embed V5 Fast Txt DZ' is $0.000088 per 1K tokens ($0.088/M). The Global meter is $0.08/M. The Microsoft Foundry catalog identifies Cohere-Embed-V5-Fast as a Direct from Azure 128K multimodal embedding model. Its separate image meter is not represented in this text-token row. |
Qwen3.7 text embedding New multilingual embedding model with configurable 2560/2048/1536/1024/768/512/256 dimensions (1024 default), up to 20 inputs per request and 128K tokens per input line. Input-only billing; no output or cached-input rate is published. | Direct API Model Studio (Intl) | 128K | $0.07CHF 0.056 | official Alibaba Cloud Model Studio's official Synchronous API page, last updated 2026-08-26 and captured 2026-08-28: the Singapore/International table lists qwen3.7-text-embedding at $0.07 per 1M input tokens, with a 128,000-token maximum per line and 20-row batches. The same page documents the model ID, dimensions and OpenAI-compatible Singapore endpoint. No output or cached-input charge is listed, so this row is input-only. |
Cohere embed-v-4-0 Best RAG / code-retrieval pick. Matryoshka dims 256/512/1024/1536 (default 1536); input_type query/document distinction. Cohere publishes this per-token rate under the 'Advanced retrieval models' tab of its pricing page; Microsoft's live Foundry meter independently reads exactly $0.12/M. | Direct API | — | $0.12CHF 0.097 | official Cohere publishes this rate first-party on cohere.com/pricing under the 'Advanced retrieval models' tab, reading verbatim '$0.12 / 1M tokens' for Embed 4 (captured 2026-08-17); the same tab also publishes an image cost of $0.47/1M, a dimension this schema doesn't model. The figure is absent from the page's server-rendered HTML — only the default 'Workplace systems' tab is server-rendered, so this tab must be clicked in a live browser to see it, and Internet Archive snapshots (SSR-only) cannot show it. |
Cohere Embed v4 Same $0.12/M as the direct lane but officially metered — currently the only published source for this rate; image embeddings bill separately at $0.47/M image tokens. | Foundry ·Global | — | $0.12CHF 0.097 | official Azure Retail Prices API 'Embed v4 Txt Glbl Tokens' at $0.00012/1K ($0.12/M) across 39 commercial regions, effective 2026-02-01, captured 2026-07-29. Companion 'Embed v4 Img Glbl Tokens' meter is $0.00047/1K ($0.47/M). US Gov regions price higher ($0.15/M) and are excluded per the commercial-majority convention. |
Cohere Embed v4 Exactly 1.1x the Global rate, the standard Data Zone premium. | Foundry ·Data Zone | — | $0.132CHF 0.106 | official Azure Retail Prices API 'Embed v4 Txt DZ Tokens' at $0.000132/1K ($0.132/M) across 21 commercial regions, effective 2025-11-01, captured 2026-07-29. Companion 'Embed v4 Img DZ Tokens' meter is $0.000517/1K ($0.517/M). US Gov regions ($0.165/M) excluded. |
OpenAI text-embedding-3-large | Foundry ·Global | — | $0.13CHF 0.105 | official Azure Retail Prices API 'text-embedding-3-large-glbl Tokens' meter: $0.00013/1K ($0.13/M) across 17 commercial regions, effective 2024-06-01, captured 2026-08-07. developers.openai.com/api/docs/pricing no longer lists per-token embeddings rates. Corroborated by developers.openai.com/api/docs/guides/embeddings, which quotes 9,615 pages per USD at ~800 tokens/page — inverting to exactly $0.13/M. US Gov Cloud bills this model under a separate 'text-embedding-3-large-regional Tokens' meter at $0.163/M, excluded per the commercial-majority convention. |
OpenAI text-embedding-3-small Cheapest solid option. | Foundry ·Global | — | $0.02CHF 0.016 | official Azure Retail Prices API 'text-embedding-3-small-glbl Tokens' meter: $0.00002/1K ($0.02/M) across 17 commercial regions, effective 2024-06-01, captured 2026-08-07. developers.openai.com/api/docs/pricing no longer lists per-token embeddings rates. Corroborated by developers.openai.com/api/docs/guides/embeddings, which quotes 62,500 pages per USD at ~800 tokens/page — inverting to exactly $0.02/M. US Gov Cloud bills this model under a separate 'text-embedding-3-small-regional Tokens' meter at $0.025/M, excluded per the commercial-majority convention. |
OpenAI text-embedding-ada-002 Legacy: worse than 3-small on cost and quality. Avoid. | Foundry ·Global | — | $0.10CHF 0.081 | official Azure Retail Prices API 'embedding-ada-glbl Tokens' meter: $0.0001/1K ($0.10/M) across 15 commercial regions, effective 2024-06-01, captured 2026-08-07. developers.openai.com/api/docs/pricing no longer lists per-token embeddings rates. Corroborated by developers.openai.com/api/docs/guides/embeddings, which quotes 12,500 pages per USD at ~800 tokens/page — inverting to exactly $0.10/M. US Gov Cloud bills this model under a separate 'embedding-ada-regional Tokens' meter at $0.125/M, excluded per the commercial-majority convention. |
Confidenceofficial published pagederived reconciled from billing †estimate pattern-inferred ‡
Field notes
Quirks & gotchas
Insight
Cheapest isn't best for code retrieval
text-embedding-3-small remains the budget pick at $0.02 / CHF 0.016 per M. Cohere Embed 5 Pro targets quality-critical retrieval at $0.12/M, while Embed 5 Fast costs $0.08/M for latency-sensitive and high-volume queries; both share an embedding space, so teams can index with Pro and query with Fast. ada-002 ($0.10 / CHF 0.081) is legacy and trails the small model on price and quality.
Note
A reference retrieval stack
For code indexing plus RAG: Cohere Embed 5 Pro for indexing, Embed 5 Fast for lower-latency queries, a vector store (LanceDB), then a strong coding LLM (DeepSeek V4 Pro, Kimi K2.7 Code, or GLM-5.2) over the retrieved chunks.
Watch out
Foundry Data Zone for embeddings is also two prices
The rows above are OpenAI's Global rate. On Microsoft Foundry, text-embedding-3-large and text-embedding-3-small each carry two Data Zone prices, not one: US/EU regions bill 1.10x Global ($0.143/M and $0.022/M), APAC regions (australiaeast, centralindia, eastasia, japaneast, japanwest, jioindiawest, koreacentral, southeastasia, southindia) bill 1.20x Global ($0.156/M and $0.024/M). Same split seen across Microsoft's first-party OpenAI lines; captured from the Azure Retail Prices API on 2026-07-27.
Other providers