StormEval
  1. StormEval
  2. Models
  3. GPT-6 Luna

Daily test results · Codex · effort medium

Is GPT-6 Luna getting worse?

Building baseline, day 3 of 14

No verdict yet: day 3 of a 14-day baseline.

Runs through
Codex
Effort
medium
Provider
codex-cli
Model id
gpt-6-luna
Tasks
New issues in real code
Updated
2026-10-21 04:41 UTC

Rating over time

Higher is better. The band is one standard error either side. The dashed line is the first day. Movement inside the band means nothing.

Was GPT-6 Luna nerfed?

No verdict yet

This is day 3 of the 14-day baseline for GPT-6 Luna. The baseline began on 7 October 2026. If the model runs every day, the baseline is complete on 20 October 2026 and the first verdict is expected on 21 October 2026.

It is too early to say either way. Until the baseline is complete there is nothing to compare a new day with, and a single day moves by chance.

Rating
1640±199 · one standard error
Change since day 1
+220from 1420 on 7 October 2026
Pass rate, latest graded day
100%latest graded day · 3 graded tasks
Days of data
3since 7 October 2026

What the runs took

Median time per task
2m 13smean 3m 06s · 0 timeouts
Output tokens per task
5.3k26% thinking · 1.22M in
Output tokens per second
34median · mean 37
Estimated cost per task
$0.021at API list prices · $0.106 a day

Estimated cost is tokens at API list prices, not money spent. A dash means no list price is set. Graded counts cover 15 tasks over 3 days; time, tokens and cost can also include finished runs excluded from grading.

Time per task

How long each task took. A task not finished in 15 minutes is a fail, and counts as 15 minutes of time spent.

  • GPT-6 Lunamedian 2m 13sslowest 12m 31s · pull-request

By task family

Pass rate, mean time and mean output tokens for each kind of problem.

Modelpull-request
GPT-6 Luna80%3m 06s5.3k

Day by day

DayRatingPassedTimeoutsTime spentMedian timeOutput tokensOutput tokens / sEst. cost
9 Oct1640 ±1993/305m 28s1m 15s11k44$0.042
8 Oct1611 ±2033/303m 46s53s13k56$0.038
7 Oct1420 ±1176/9037m 15s3m 18s56k25$0.238

How this is measured

GPT-6 Luna gets the same new tasks as every other model. Each task is an issue in a large, working codebase, and the model writes the change that resolves it with its own coding tool, as it would for a pull request. A program runs the tests to check the change. Only one run per task counts toward this model's results.

The model is compared with itself: the last 7 days against its first 14 days. A verdict is only given when the difference is too large to be chance. A task not finished in 15 minutes counts as a fail.

Read the full method

Not a leaderboard. Models run at different effort settings and through different tools.

Other models