Same questions. Three ways to answer.

20 rows · click any column to sort · expand a row for league breakdown

	#	Model	SetupHow the model got its answer: Blitz agent (sports stack), DB tools (generic SQL on the warehouse, no Blitz), or Web (browse only, no DB).	ThinkThinking budget. Low is fast; Max is slow. Blitz = Agent (tools).	Overall55% Correct + 25% Speed + 20% Cost.	CorrectShare of sports questions answered correctly (0–100).	SpeedFrom average response time. Higher = faster.	CostFrom average $/question. Higher = cheaper.	TimeAverage seconds per question.	$/QAverage estimated cost per question.
	1	Blitz Blitz	Blitz	Agent	96.5	98.7	92.9	94.7	7.6s	$0.016
	2	Claude Sonnet 5 Anthropic	DB tools	High	66.5	54.3	78.9	84.8	20.2s	$0.046
	3	Gemini 3.1 Pro Google	DB tools	High	64.7	49.3	81.2	86.5	18.1s	$0.041
	4	Claude Opus 5.5 Anthropic	DB tools	High	63.8	57.2	72.2	71.2	26.2s	$0.086
	5	GPT-5.6 Sol OpenAI	DB tools	High	63.7	51.3	76.7	81.4	22.2s	$0.056
	6	Grok 4 xAI	DB tools	High	63.5	47.4	80.1	87.1	19.1s	$0.039
	7	Claude Sonnet 5 Anthropic	DB tools	Low	63.1	35.5	96.9	96.6	4.0s	$0.010
	8	GPT-5.6 Sol OpenAI	DB tools	Low	61.9	33.5	96.6	96.3	4.2s	$0.011
	9	Gemini 3 Flash Google	DB tools	Low	60.8	29.6	98.9	99.2	2.2s	$0.003
	10	DeepSeek V4 Flash DeepSeek	DB tools	Low	59.9	27.6	99.2	99.5	1.9s	$0.002
	11	OpenAI o3 OpenAI	DB tools	High	59.3	53.3	65.5	67.9	32.2s	$0.096
	12	Claude Opus 5.5 Anthropic	DB tools	Max	55.7	64.1	39.8	52.6	55.4s	$0.142
	13	Claude Sonnet 5 Anthropic	Web	Low	47.3	23.7	67.7	86.5	30.2s	$0.041
	14	Gemini 3 Flash Google	Web	Low	47.0	19.7	72.2	90.5	26.2s	$0.028
	15	GPT-5.6 Sol OpenAI	Web	Low	45.3	21.7	65.5	84.8	32.2s	$0.046
	16	Claude Sonnet 5 Anthropic	Web	High	40.2	35.5	28.6	67.9	65.5s	$0.096
	17	Gemini 3.1 Pro Google	Web	High	40.2	31.6	34.2	71.2	60.4s	$0.086
	18	GPT-5.6 Sol OpenAI	Web	High	36.7	33.5	23.0	62.8	70.5s	$0.112
	19	Claude Opus 5.5 Anthropic	Web	High	34.5	39.5	11.8	49.3	80.6s	$0.152
	20	OpenAI o3 OpenAI	Web	High	24.4	32.6	0.0	32.3	100.8s	$0.203

Setup is the answer path, not the model brand: Blitz = sports agent stack; DB tools = frontier model + generic SQL on the same warehouse; Web = browse only. Sort by Correct for the sports-agent story; sort by Speed/Time for flash models. Blitz is measured; DB/Web rows are literature-calibrated placeholders.