Head to head
Gemini 3.1 Pro (preview) vs GPT-5.4
Gemini 3.1 Pro (preview): currentGPT-5.4: currentheavy tier
Verdict · modelled basis
GPT-5.4 costs 1.0x less per solved task.
On output token price alone, Gemini 3.1 Pro (preview) looks 1.3x cheaper (1.3x on input). The ranking reverses once you divide by pass rate — the cheaper tokens do not win the task.
Source: Scale SEAL leaderboard, SWE-bench Pro · verified 2026-08-21 · prices from each provider's pricing page
Input $/M tokens
lower is better
Output $/M tokens
lower is better
Cost per solved task
modelled basis · lower is better
| Gemini 3.1 Pro (preview) | GPT-5.4 | |
|---|---|---|
| Input $/Mtok | $2 | $2.5 |
| Output $/Mtok | $12 | $15 |
| Cached input $/Mtok | $0.2 | $0.25 |
| Pass rate | 46% | 59% |
| Pass rate source | SEAL | SEAL |
| Cost basis | modelled | modelled |
| Cost per solved task | $4.30 | $4.19 |
verified 2026-08-21 · heavy task tier; measured rows ignore the tier· Methodology· Gemini 3.1 Pro (preview)· GPT-5.4