Measured by Spring Prompt

BulletBench

When every second of thinking time comes off the clock, which models are fast enough to still make good decisions?

Results dated: 29 Sep 2026 to 30 Sep 2026
Models: 16
Unit: ladder Elo
Licence: Spring Prompt original

BulletBench 180+2: ladder Elo, ladder Elo, higher is better
#	Model	Ladder Elo · 95% range ladder Elo, higher is better	Median move time milliseconds
1	Gemini 3.8 FlashGoogle	1,126 949–1,347	4.0 s
1	GPT-6 AstraOpenAI	839 595–1,072	5.6 s
1	Gemini 3.1 Pro PreviewGoogle	683 487–1,130	6.2 s
2	Gemini 3.5 Flash LiteGoogle	707 458–933	0.8 s
2	Claude Opus 5.5Anthropic	557 370–727	7.5 s
2	Claude Sonnet 5.5Anthropic	557 370–727	8.1 s
2	GPT-6 LunaOpenAI	557 370–727	7.0 s
2	Mistral Medium 3.5Mistral	555 396–765	0.5 s
2	Claude Haiku 4.5Anthropic	491 299–659	1.3 s
2	Claude Fable 5.1Anthropic	358 0–619	5.4 s
6	Grok 4.7xAI	147 0–387	10.1 s
12	Qwen3.8 Max (0902)Alibaba	0	7.4 s
12	DeepSeek V4 Pro 0423DeepSeek	0	6.4 s
12	Muse Spark 1.3Meta	0	20.7 s
12	Kimi K3Moonshot AI	0	6.8 s
12	GPT-6 SolOpenAI	0	8.2 s

Models share a rank when their ranges overlap. Each model runs at its provider's default reasoning setting. Some providers think at length by default and others barely at all, so this is what you get without tuning.

What it measures

Decision quality under real time pressure
Response latency at the provider's default settings

What it does not measure

Any business skill: this is chess against an engine
Chess strength without a clock

Method

Games against a Stockfish ladder at 3 minutes plus 2 seconds a move
Every second of response time comes off the model's clock
Ladder Elo with a 95% interval; it orders models, it is not a FIDE rating