Prepaid
Pay as you goAdd balance when needed. Funds are reserved before dispatch and settled from terminal usage.
Independent price check · 2026-08-31
Compare API costs before you ship.
A practical comparison of prepaid blended pricing, official split tariffs and subscription-style access. Start with your real token mix, not a headline number.
Model.sale public rates
Prices are per 1M charged tokens. The initial reservation is temporary and released after terminal usage; it is not a per-request fee. Prices remain listed when a model is temporarily unavailable.
| Model | Blended / 1M | Initial reservation | Status |
|---|---|---|---|
| claude-haiku-4-5 | $0.075 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| claude-opus-4-5 | $0.30 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| claude-opus-4-6 | $0.40 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| claude-opus-4-7 | $0.50 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| claude-opus-4-8 | $0.60 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| claude-sonnet-4-5 | $0.15 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| claude-sonnet-4-6 | $0.25 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| deepseek-v4-flash | $0.20 | $0.01 (released) | LIVE |
| deepseek-v4-pro | $0.20 | $0.01 (released) | LIVE |
| glm-5 | $0.12 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| glm-5.1 | $0.16 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| glm-5.2 | $0.20 | $0.01 (released) | LIVE |
| glm-5.3 | $0.20 | $0.01 (released) | LIVE |
| glm-5.3-flash | $0.10 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| gpt-5.4 | $0.10 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| gpt-5.4-20 | $0.10 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| gpt-5.4-mini | $0.09 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| gpt-5.5 | $0.45 | $0.01 (released) | LIVE |
| gpt-5.6-luna | $0.09 | $0.01 (released) | LIVE |
| gpt-5.6-sol | $0.45 | $0.01 (released) | LIVE |
| gpt-5.6-terra | $0.15 | $0.01 (released) | LIVE |
| gpt-6-astra | $1.125 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| gpt-image-2 | $0.10 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| kimi-k2.5 | $0.07 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| kimi-k2.6 | $0.11 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| kimi-k3 | $0.24 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| mimo-v2-omni | $0.05 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| mimo-v2-pro | $0.05 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| mimo-v2.5 | $0.01 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| mimo-v2.5-pro | $0.03 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| minimax-m2.5 | $0.02 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| minimax-m2.7 | $0.04 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| minimax-m3 | $0.04 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| qwen-3.8 | $0.25 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| qwen3.5-plus | $0.05 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| qwen3.6-plus | $0.04 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| qwen3.7-max | $0.11 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
| qwen3.7-plus | $0.03 | $0.01 (released) | TEMPORARILY UNAVAILABLE |
Official APIs
Usually split ratesInput, cached input and output can have different prices and context tiers.
Exact usage
Measured settlementA temporary reservation protects the wallet, then the final charge uses only terminal tokens returned by the API.
Popular API alternatives
Model.sale is a prepaid single-service catalog. The products below are useful alternatives, but they are not identical offers: some are multi-provider routers, some are direct open-model platforms, and some bill after usage. We compare the buying model and the documented trade-offs, not model quality or a universal “cheapest” claim.
| Service | Billing | Catalog / routing | Best fit | Checked |
|---|---|---|---|---|
| OpenRouter Multi-provider aggregator | Pay-as-you-go credits; 5.5% credit purchase fee (minimum $0.80); crypto purchases have a 5% fee. | 400+ models and 70+ providers on the pay-as-you-go plan. Provider selection, fallbacks and auto-routing are available. | Teams that want breadth and provider choice in one account. Trade-off: More routing and pricing dimensions to understand; a headline per-token price is not a blended workload cost. | 2026-08-31 · source |
| Together AI Open-model inference platform | Serverless inference is pay-per-token with no provisioning cost or minimum; dedicated options use reserved capacity. | Open-weight models including DeepSeek, Qwen, Kimi, GLM and Llama families. You select the model/endpoint; serverless capacity is shared and best-effort. | Developers who need open models and want a direct, programmable API. Trade-off: You still manage model selection, split input/output math and the provider account; dedicated capacity adds a different cost model. | 2026-08-31 · source |
| Fireworks AI Inference platform | Serverless inference is per-token; dedicated deployments are billed by GPU time. New users may receive free credits. | A curated serverless catalog focused on open models, with custom and dedicated deployment paths. You choose a serverless model or run a dedicated deployment; no universal multi-provider router is promised. | Fast prototypes that may later need a dedicated open-model deployment. Trade-off: GPU-hour costs and operational choices matter once you leave serverless; compare total workload cost, not only token price. | 2026-08-31 · source |
| Groq Low-latency inference platform | Developer usage is billed in arrears with progressive thresholds; the free/developer limits and payment requirements differ. | A focused catalog of production and preview models, published with model-specific limits and token rates. You select a Groq model; the platform emphasizes predictable low-latency serving rather than cross-provider routing. | Latency-sensitive applications using the models in Groq's active catalog. Trade-off: It is not a broad aggregator, and postpaid billing is a different budget model from a prepaid wallet. | 2026-08-31 · source |
Source pages and checked dates are part of the comparison. Terms, fees, model catalogs and limits can change; verify the linked documentation before making a production decision. Model.sale does not imply affiliation with any listed service.
Illustrative workload examples
These examples compare billing shapes for the same 800K input + 200K output mix. They do not claim that similarly named models have identical quality or behavior.
| Reference family | Model.sale blended | Model.sale example | Official split example |
|---|---|---|---|
| GPT-5.6 Luna | $0.09/M | $0.09 | $0.40 · $0.20 input + $1.20 output |
| GPT-5.6 Terra | $0.15/M | $0.15 | $4.00 · $2 input + $12 output |
| GPT-5.6 Sol | $0.45/M | $0.45 | $7.20 · $4 input + $20 output |
The Model.sale column applies one blended rate to 1M charged tokens. The official column applies the published input/output rates to the stated mix. Change the mix and the result changes.
Prepaid vs subscription access
| Question | Prepaid API | Subscription product |
|---|---|---|
| How you pay | Deposit balance, then pay usage | Recurring seat or plan fee |
| Best for | Variable workloads, scripts and coding agents | Predictable seats and bundled product features |
| Cost visibility | Request-level token and ledger records | Usually plan usage limits or quotas |
| Risk to budget | Set key and daily spend limits | Recurring charge until cancelled |
How to calculate your effective cost
- Export a representative hour or day of input, cached and output tokens.
- Separate retries, failed calls and streaming disconnects.
- Apply each provider’s exact input/output/cache rules and context tier.
- Apply Model.sale blended price to the measured charged-token total; the initial reservation is released after settlement.
- Compare the resulting total, latency, availability and model capability together.
Official reference rows below are informational, dated and linked to their source. They are not a promise of parity or a benchmark.
Official published rates
USD per million tokens, standard published rates. Cached input is shown separately where the source publishes it.
| Model | Input | Cached input | Output | Source |
|---|---|---|---|---|
| OpenAI GPT-5.6 Sol | $4.00 | See source | $20.00 | Official pricing |
| OpenAI GPT-5.6 Terra | $2.00 | See source | $12.00 | Official pricing |
| OpenAI GPT-5.6 Luna | $0.20 | See source | $1.20 | Official pricing |
| Anthropic Claude Opus 4.8 | $5.00 | $0.50 | $25.00 | Official pricing |
| Anthropic Claude Sonnet 4.6 | $3.00 | $0.30 | $15.00 | Official pricing |
| Anthropic Claude Haiku 4.5 | $1.00 | $0.10 | $5.00 | Official pricing |
Not an equivalence claim
Names, context limits and quality tiers can differ. Treat the table as a billing-shape reference.
Rates can change
Check the source date and verify official pages before committing to a workload.
Need exact math?
Read the pricing methodology for measured usage, reservations and settlement.