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: milliseconds
Licence: Spring Prompt original

BulletBench 180+2: median move time, milliseconds, lower is better
#	Model	Median move time milliseconds, lower is better	Ladder Elo ladder Elo
1	Mistral Medium 3.5Mistral	0.5 s	555
2	Gemini 3.5 Flash LiteGoogle	0.8 s	707
3	Claude Haiku 4.5Anthropic	1.3 s	491
4	Gemini 3.8 FlashGoogle	4.0 s	1,126
5	Claude Fable 5.1Anthropic	5.4 s	358
6	GPT-6 AstraOpenAI	5.6 s	839
7	Gemini 3.1 Pro PreviewGoogle	6.2 s	683
8	DeepSeek V4 Pro 0423DeepSeek	6.4 s	0
9	Kimi K3Moonshot AI	6.8 s	0
10	GPT-6 LunaOpenAI	7.0 s	557
11	Qwen3.8 Max (0902)Alibaba	7.4 s	0
12	Claude Opus 5.5Anthropic	7.5 s	557
13	Claude Sonnet 5.5Anthropic	8.1 s	557
14	GPT-6 SolOpenAI	8.2 s	0
15	Grok 4.7xAI	10.1 s	147
16	Muse Spark 1.3Meta	20.7 s	0

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