Temporarily unavailable
GPT-5 nano
Provided by openai
Fast, low-cost GPT model for classification and extraction baselines.
- RunBatch input rate
- $0.12500 / 1M
- RunBatch output rate
- $1.00000 / 1M
- Context window
- 400,000
- Structured output
- Yes
Transparent by design
Know the cost before you scale.
Run a representative sample, compare configurations, and review the projected full-workload cost before committing. Transparent per-model RunBatch rates remain available for direct runs.
Managed model pricing
RunBatch rates by model.
Managed eligibility pending
Managed default
GPT-5 mini
Provided by openai
Reliable standard processing for classification and structured extraction.
Available for exact configuration selection.
- RunBatch input rate
- $0.62500 / 1M
- RunBatch output rate
- $5.00000 / 1M
- Context window
- 400,000
- Structured output
- Yes
Exact configuration only
GPT-OSS 20B
Provided by groq
Fast open-weight model available for evaluated advanced workloads.
Available for exact configuration selection.
- RunBatch input rate
- $0.09375 / 1M
- RunBatch output rate
- $0.37500 / 1M
- Context window
- 131,072
- Structured output
- Yes
Exact configuration only
GPT-OSS 120B
Provided by groq
Higher-capability open-weight model for demanding extraction workloads.
Available for exact configuration selection.
- RunBatch input rate
- $0.18750 / 1M
- RunBatch output rate
- $0.75000 / 1M
- Context window
- 131,072
- Structured output
- Yes
Temporarily unavailable
Qwen 3.8 27B
Provided by groq
Fast Qwen-family model for structured and multilingual races.
- RunBatch input rate
- $1.00000 / 1M
- RunBatch output rate
- $5.00000 / 1M
- Context window
- 131,072
- Structured output
- Yes
Managed eligibility pending
DeepSeek V4 Flash
Provided by fireworks
DeepSeek-family batch inference on Fireworks.
- RunBatch input rate
- $0.55000 / 1M
- RunBatch output rate
- $1.65000 / 1M
- Context window
- 1,048,576
- Structured output
- Yes
Temporarily unavailable
GPT-OSS 120B
Provided by fireworks
GPT-OSS baseline for cross-provider comparisons.
- RunBatch input rate
- $0.37500 / 1M
- RunBatch output rate
- $1.50000 / 1M
- Context window
- 131,072
- Structured output
- Yes
Exact configuration only
Llama 3.3 70B Turbo
Provided by together
Llama-family discounted batch inference on Together.
Available for exact configuration selection.
- RunBatch input rate
- $2.60000 / 1M
- RunBatch output rate
- $2.60000 / 1M
- Context window
- 131,072
- Structured output
- Yes
Managed eligibility pending
Qwen 3.5 9B
Provided by together
Current Qwen-family model with structured output support and strong price-performance.
Available for exact configuration selection.
- RunBatch input rate
- $0.42500 / 1M
- RunBatch output rate
- $0.62500 / 1M
- Context window
- 262,144
- Structured output
- Yes
Exact configuration only
GPT-OSS 20B
Provided by together
GPT-OSS baseline for cross-provider comparisons.
Available for exact configuration selection.
- RunBatch input rate
- $0.12500 / 1M
- RunBatch output rate
- $0.50000 / 1M
- Context window
- 131,072
- Structured output
- Yes
All rates are in USD per 1 million tokens. These are RunBatch retail rates for the named model, not the model provider’s API prices. The selected provider, model, and rates are saved with your batch and do not change during execution. Token rates are only one part of workload cost; Races compare observed execution on your data.
How billing works
- Estimate for free. We inspect every dataset row. The estimate uses approximate input tokens and your maximum output tokens per row; it is not a promised final price.
- Reserve a maximum budget. The review screen shows the exact amount held from your balance, including a conservative safety buffer. It is not a card payment or a final charge.
- Pay for consumed tokens. Final charge = billable input tokens × input rate + billable output tokens × output rate, divided by 1 million, rounded up to whole cents. If any tokens are consumed, a $0.50 minimum applies once per batch.
- Get unused balance back. Your charge never exceeds the confirmed reservation. If no tokens are consumed, the charge is $0. Unused funds are released after usage is verified; missing usage pauses settlement for review.
What counts as usage?
We use the provider-reported token counts for each processed request, including prompt instructions and schema overhead. Input includes cached tokens at the same published input rate. Output includes reasoning tokens when included in the provider’s output usage.
Provider-reported tokens are billable even when an answer is truncated, refused, fails output validation or belongs to a transient attempt. Requests rejected without token consumption cost $0. Automatic retries cover only reconciled transient provider failures, stay within the confirmed attempt budget, and include only unfinished rows. The managed default uses only configurations that meet RunBatch’s current production qualification; exact configurations remain available for your own evaluations.
The output-token setting is a per-row limit, not a minimum purchase. A shorter answer can reduce the charge, down to the batch minimum.
Previously issued usage-billed and fixed-price batches retain their original terms.
Example · GPT-5 nano
If the provider reports 100,000 input tokens and 100,000 output tokens: input $0.012500 + output $0.100000 = $0.112500. After the minimum and rounding, the final charge is $0.50. Any reservation above this is released. This is an illustration, not a prediction of your batch’s usage.
Your balance lives in Billing.
Add funds and track reservations and charges in your account. Minimum top-up: $10. No subscription and no automatic recharge.