Project Peak
← Leaderboard
Hosted medium reasoning window 400K $0.25/Mtok input

GPT-5 mini

At the Frontier difficulty, GPT-5 mini stays reliable to 29K of its 400K advertised window for critical use (7% of the window), at 1.3M /$ of reliable context.

Frontier difficulty

Hardest tasks, tuned to bend the strongest models

baseline 100% · value 1.3M /$
0 reliable to 29K window 400K
Critical ≥95%
29K *
CI 2.9K to 65K
Right 95% of the time. Use it where a wrong answer is costly.
Normal ≥85%
63K
CI 3.5K to 192K
Right 85% of the time. A person reviews the output.
Exploratory ≥75%
165K
CI 25K to 262K
Right 75% of the time. Fine for drafts and ideation.
Reliability as context grows
2.9K
4.1K
8.2K
16K
33K
66K
131K
262K

Standard difficulty

A moderate task difficulty

baseline 100% · value 3.7M /$
0 reliable to 262K+ window 400K
Critical ≥95%
262K+ *
Right 95% of the time. Use it where a wrong answer is costly.
Normal ≥85%
262K+ *
Right 85% of the time. A person reviews the output.
Exploratory ≥75%
262K+
Right 75% of the time. Fine for drafts and ideation.
Reliability as context grows
2.0K
2.9K
4.1K
8.2K
16K
33K
66K
131K
262K

Light difficulty

Lighter tasks, where cheaper and self-hosted models still bend in window

baseline 100% · value 3.8M /$
0 reliable to 262K+ window 400K
Critical ≥95%
262K+ *
Right 95% of the time. Use it where a wrong answer is costly.
Normal ≥85%
262K+ *
Right 85% of the time. A person reviews the output.
Exploratory ≥75%
262K+
Right 75% of the time. Fine for drafts and ideation.
Reliability as context grows
2.0K
2.9K
4.1K
8.2K
16K
33K
66K
131K
262K

Usable context is the longest context a model stays at or above a usage bar, read from a fit that borrows strength across the difficulty tiers so the higher bars resolve without deep per-cell sampling. A star marks a bar whose lower confidence bound is not resolved at this depth. Value is reliable context per dollar of input price; self-hosted models are priced by amortized GPU-hours.

Secondary metric: difficulty ceiling

Usable context above is the primary measure. As a secondary, difficulty-axis view (how hard a coupled reasoning task it sustains as context grows), GPT-5 mini scores 66/100.