Model pricing: one subscription vs per-token APIs
There are two different ways to pay for access to AI models. A vendor API bills for the tokens processed by individual requests. Borg groups the live models in this directory under one subscription. The prices below show vendor API list rates for comparison, not what Borg charges per request.
How API pricing works
An API request sends input tokens, such as instructions and documents, and receives output tokens generated by the model. Vendors often publish separate prices per million input tokens and per million output tokens. Cached tokens can have another rate when a vendor supports caching and a request qualifies. A low input rate does not necessarily make a model inexpensive for a task that produces a lot of output. Your total API bill depends on the model, the amount of text sent and generated, cache eligibility, and the vendor's current terms. Each model page lists the vendor list price when we have one, with a link to its source. Unknown rates are not estimates.
How Borg pricing works
Borg provides the models in this catalog through a flat subscription instead of charging the listed vendor rate per token on every request. Every live model shown in the table is included in the catalog. You can choose a lane directly, or use Borg Fusion to pick a model for each step of a task. Routing can change which model handles a step, so a single workflow need not rely on one vendor's API throughout. For current subscription terms, see Borg pricing; for organizational use, visit Borg for Business. The vendor API amounts below are not subscription prices, and they should not be added to a Borg bill.
Vendor API list prices by model
These rates are in US dollars per million tokens. The input and output columns are separate: compare the mix your actual work produces rather than treating them as one rate. Follow a model link for the vendor price source and details about that model's published specifications. Not published means we have no cited list price here, not that the model is free to use through a vendor API.
| Model | Vendor input / 1M tokens | Vendor output / 1M tokens | Source and conditions |
|---|---|---|---|
| Claude Fable 5.1 | $10 | $50 | Vendor source |
| Claude Opus 5.5 | $4 | $20 | Vendor source |
| Claude Sonnet 5.5 | $2 | $10 | Vendor source |
| Claude Haiku 4.5 | $1 | $5 | Vendor source |
| Claude Opus 4.6 | $5 | $25 | Vendor source |
| Claude Sonnet 4.6 | $3 | $15 | Vendor source |
| GPT-5.5 | $5 | $30 | Vendor source Standard global rates for prompts up to 272,000 input tokens. Above 272,000 input tokens, input and cached-input rates are 2× and output rates are 1.5× for the full session. |
| GPT-5.6 Luna | $0.2 | $1.2 | Vendor source Standard global rates for prompts up to 272,000 input tokens. Above 272,000 input tokens, input and cached-input rates are 2× and output rates are 1.5× for the full request. |
| GPT-5.6 Sol | $4 | $20 | Vendor source Standard global rates for prompts up to 272,000 input tokens. Above 272,000 input tokens, input and cached-input rates are 2× and output rates are 1.5× for the full request. Promotional pricing is guaranteed at least through November 21, 2026; a later rate has not been announced. |
| GPT-5.6 Terra | $2 | $12 | Vendor source Standard global rates for prompts up to 272,000 input tokens. Above 272,000 input tokens, input and cached-input rates are 2× and output rates are 1.5× for the full request. |
| GPT-6 Astra | $10 | $50 | Vendor source Standard global rates for prompts up to 272,000 input tokens. Above 272,000 input tokens, input and cached-input rates are 2× and output rates are 1.5× for the full request. |
| GPT-6 Luna | $0.1 | $0.5 | Vendor source Standard global rates for prompts up to 272,000 input tokens. Above 272,000 input tokens, input and cached-input rates are 2× and output rates are 1.5× for the full request. |
| GPT-6 Sol | $2 | $10 | Vendor source Standard global rates for prompts up to 272,000 input tokens. Above 272,000 input tokens, input and cached-input rates are 2× and output rates are 1.5× for the full request. |
| GPT-6.1 Sol | $2 | $10 | Vendor source Standard global rates for prompts up to 272,000 input tokens. Above 272,000 input tokens, input and cached-input rates are 2× and output rates are 1.5× for the full request. |
| gpt-oss-120b | Not published | Not published | Not published |
| Gemini 2.5 Flash | $0.3 | $2.5 | Vendor source Input and cache rates shown are for text/image/video, not audio. |
| Gemini 2.5 Flash-Lite | $0.1 | $0.4 | Vendor source Input and cache rates shown are for text/image/video, not audio. |
| Gemini 2.5 Pro | $1.25 | $10 | Vendor source Rates for prompts of at most 200,000 tokens; longer prompts cost more. |
| Gemini 3 Flash | $0.5 | $3 | Vendor source Input and cache rates shown are for text/image/video, not audio. |
| Gemini 3.1 Flash Image | $1.5 | $7.5 | Vendor source Output is text/thinking; image output costs $30 per million tokens. |
| Gemini 3.1 Flash-Lite | $0.25 | $1.5 | Vendor source Input and cache rates shown are for text/image/video, not audio. |
| Gemini 3.1 Pro | $2 | $12 | Vendor source Rates for prompts of at most 200,000 tokens; longer prompts cost more. |
| Gemini 3.5 Flash | $1.5 | $9 | Vendor source |
| Gemini 3.5 Flash-Lite | $0.3 | $2.5 | Vendor source |
| Gemini 3.6 Flash | $0.75 | $3.75 | Vendor source Introductory rates through December 31, 2026. From January 1, 2027: $1.50 input / $7.50 output / $0.15 cached input per 1M tokens. |
| Gemini 3.7 Flash | $0.75 | $3.75 | Vendor source Introductory rates through December 31, 2026. From January 1, 2027: $1.50 input / $7.50 output / $0.15 cached input per 1M tokens. |
| Gemini 3.8 Flash | $0.75 | $3.75 | Vendor source Introductory rates through December 31, 2026. From January 1, 2027: $1.50 input / $7.50 output / $0.15 cached input per 1M tokens. |
| Grok 4.3 | $1.25 | $2.5 | Vendor source Global rates for prompts below 200,000 input tokens. At 200,000 input tokens or more, all tokens use long-context rates: $2.5 input / $5 output / $0.4 cached input per 1M tokens. US regional inference costs 1.1× the global rate. |
| Grok 4.5 | $2 | $6 | Vendor source Global rates for prompts below 200,000 input tokens. At 200,000 input tokens or more, all tokens use long-context rates: $4 input / $12 output / $0.6 cached input per 1M tokens. US regional inference costs 1.1× the global rate. |
| Grok 4.6 | $2 | $6 | Vendor source Global rates for prompts below 200,000 input tokens. At 200,000 input tokens or more, all tokens use long-context rates: $4 input / $12 output / $1 cached input per 1M tokens. US regional inference costs 1.1× the global rate. |
| Grok 4.7 | $2 | $6 | Vendor source Global rates for prompts below 200,000 input tokens. At 200,000 input tokens or more, all tokens use long-context rates: $4 input / $12 output / $1 cached input per 1M tokens. US regional inference costs 1.1× the global rate. |
| Kimi K3 | $3 | $15 | Vendor source |
| Kimi K2.7 Code | $0.95 | $4 | Vendor source |
| GLM-4.5 | $0.6 | $2.2 | Vendor source |
| GLM-4.5-Air | $0.2 | $1.1 | Vendor source |
| GLM-4.6 | $0.6 | $2.2 | Vendor source |
| GLM-4.7 | $0.6 | $2.2 | Vendor source |
| GLM-5 | $1 | $3.2 | Vendor source |
| GLM-5-Turbo | Not published | Not published | Not published |
| GLM-5.1 | $1.4 | $4.4 | Vendor source |
| GLM-5.2 | $1.4 | $4.4 | Vendor source |
| GLM-5.3 | $1.4 | $4.4 | Vendor source |
| Muse Spark 1.1 | $1.25 | $4.25 | Vendor source |
| Muse Spark 1.2 | $1.25 | $4.25 | Vendor source |
| Muse Spark 1.3 | $1.25 | $4.25 | Vendor source |
| Borg Fusion | Not published | Not published | Not published |
| Borg Fusion Coding | Not published | Not published | Not published |
| Borg Fusion Design | Not published | Not published | Not published |
| Borg Fusion Fast | Not published | Not published | Not published |
| Atlas Max | Not published | Not published | Not published |
Claude
GPT
Gemini
Grok
KimiWhich approach fits your usage?
A per-token API may be cheaper when you make very few requests, especially if the inputs and outputs are small. In that case a usage-based bill can be less than paying for a subscription you rarely use. Compare your likely token volumes and the vendor's actual rates before deciding; there is no universal break-even point. A subscription may be cheaper for daily use, repeated experiments, or teams who regularly switch between models. It can also make budgeting simpler when the size of individual requests varies. Use the Cost Savings Calculator to test your own usage scenario rather than assuming one model or billing method always wins. Check the current Borg plans and each vendor's current terms before making a purchase decision.
FAQ
Does Borg charge the vendor API rate for each request?
No. Borg provides access to the models in this directory through a subscription. The vendor rates in the table are list prices for using vendor APIs, not additional charges from Borg. Check the current Borg plan for subscription details.
Why do some vendor prices say Not published?
A missing price means this catalog does not have a cited vendor list price for that model. It does not mean the vendor API is free. Open the model page for its available sources and check the vendor before planning an API budget.
Can I choose a model instead of using Borg Fusion?
Yes. Each model page lists its Borg lane ids for direct selection. Borg Fusion is the alternative when you want routing to choose a model for each step rather than specifying a lane yourself.