Skip to content
RCreddit.com·

Astra Low vs Sol High vs Terra High credit usage measurement

AI summary

A developer conducted a test comparing the credit usage of GPT-6 Astra Low, GPT-5.6 Sol High, and GPT-5.6 Terra High on an identical Codex task. The initial test with Astra Low yielded a cost of $2.23. To ensure accuracy and rule out cached token effects, the Astra Low test was re-run, producing the exact same cost of $2.23, confirming consistent measurement for the comparison of these models.

Time & source

Times shown in UTC

Display time zone: UTC

Local time zone unavailable; showing UTC.

PublishedOffset at this time: UTC+0Sep 11, 2026, 12:46 UTC

IngestedOffset at this time: UTC+0Sep 12, 2026, 12:01 UTC

Published
Sep 11, 2026, 12:46
Ingested
Sep 12, 2026, 12:01
Source type
Dev community
Tier
Community
Source status
Healthy

Tier is a per-source editorial setting, not a per-item score.

Discussion trend

No comparison yet
Latest 24h versus previous 24h snapshot means · 7-day curve

The percentage is based on collected discussion signal, not new comments or independent people. The curve only compares the same topic across time.

Article

I ran a small test comparing the cost of GPT-6 Astra Low, GPT-5.6 Sol High, and GPT-5.6 Terra High on the exact same Codex task.

I asked all three to export the full conversation history from a medium-sized Codex chat as an .md file, using the least amount of tokens possible. I used the exact same prompt for each model.

I recorded my account balance before and after each task, then calculated the actual cost of each individual run from the difference.

Cost per completed task:

- Astra Low: $2.23

- Sol High: $0.32

- Terra High: $0.42

Yeah... I was very surprised by the difference between Astra and the other models. Astra Low cost roughly 6.9× as much as Sol High for the exact same task.

Obviously, this is a tiny test, so the results may or may not be representative of broader real-world usage. But it is a measurement nonetheless.

Clarification: Same session, same task. Changed the model, "edited" the last message, sent it again.

Edit: To check whether running the same task was using cached tokens and therefore lowering the cost of subsequent tests, I ran the first test (Astra Low) again. The result was exactly the same, down to the last cent: $2.23.

Source·reddit.com