GPT-6.2 Astra and Sol Pricing: Cost for 100 Tasks

Source: Elser AI

For 100 API tasks with 10,000 uncached input tokens and 2,000 billable output tokens each, GPT-6.2 Astra costs an estimated $16 and GPT-6.2 Sol costs $4. The example uses Standard short-context rates and excludes tool costs, retries, cache activity, and other surcharges. Every figure below is a calculation from the published rates. Astra API pricing · Sol API pricing

The calculation for 100 tasks

| Standard short-context rate per million tokens | Astra | Sol | |---|---:|---:| | Uncached input | $8.00 | $2.00 | | Cached input reads | $0.40 | $0.10 | | Cache writes, 30-minute retention | $10.00 | $2.50 | | Output | $40.00 | $10.00 |

Across 100 requests, the example sends 1 million input tokens and uses 200,000 output tokens. Astra’s input costs $8 and its output costs $8. Sol’s input costs $2 and its output costs $2. When estimating from a pilot, use the billable output count reported by the API. Reasoning tokens can affect the total even when the answer displayed to the reader is short. OpenAI API pricing

What changes if you use both models?

Using Sol for 80 of these requests and Astra for 20 would cost $6.40 if each task goes to exactly one model. If every task runs through Sol first and 20 are then rerun on Astra, the estimate rises to $7.20. Are those reruns worth paying for? Compare the additional cost with the reviewer time they save and the number of drafts that pass your acceptance checks.

For a team repurposing an approved script, an initial pilot might use Sol for narrative variations and Astra for cases where a revision has to reconcile competing constraints. Add columns for tokens, tool charges, retries, and accepted outputs. Dividing the full spend by accepted outputs accounts for failed attempts.

Where estimates usually drift

Repeated material. Eligible reusable prefixes can incur a cache write charge and then a lower read rate on later requests. A cache write and a cache read are separate cost categories. Use reported cache usage and the retention period in your estimate. Prompt caching guide

Long inputs. For requests above 272K input tokens, the model pages document twice the short-context input and cache rates and 1.5 times the output rate for the full request. The threshold applies per request, so 100 small requests do not become long-context requests just because their combined usage exceeds it. Astra rates · Sol rates

Speed and scheduling. Fast, Batch, Ultrafast, and eligible background completion options have separate terms. Determine the service tier actually used before treating the Standard-rate estimate as a bill forecast. API pricing

Frequently asked questions

Does a ChatGPT Work subscription use the same price table? The table is for API billing. Work and Codex subscription usage have their own rules and allowances. ChatGPT Work and Codex pricing

What should I budget for first? Use a handful of real requests to measure billable tokens and tool use, then multiply by expected volume and add the cost of observed retries.

Latest Posts