C:\MS-DOS Prompt — beckon

C:\> beckon run bench --tests 8

The vibe coder's benchmark

Eight one-shot tests, identical conditions. Every prompt, artifact, and vote public.

Runs live inside Beckon

Season score

48/ 50 pts

gpt-5.6 Sol Ultra leads the season

Arena ELO

1065· 8W–2L

claude-fable-5 Max effort tops the AI-judge ladder

Token efficiency

2.2ktoks / point

gpt-5.6 Sol Ultra does the most with the least

❯ Scoreboardper test

Runs first try · polish · prompt adherence · wow factor — ten points per test, ties to the cheaper run. ◌ run captured, no score. New runs are decided by the People's Vote and the AI panel.

48gpt-5.6 Sol Ultra 34claude-fable-5 Max effort 10kimi-code/k3 10claude-opus-4-8

Model	Game dev	Creative coding		UI & design		Agentic		Debugging	Total
Model	01	02	04	03	05	06	08	07	Total
#1 gpt-5.6 Sol Ultra OpenAI	8	10	10	·	·	10	10	·	48/50
#2 claude-fable-5 Max effort Anthropic	4	6	4	·	·	10	10	·	34/50
#3 kimi-code/k3 Moonshot	·	10	·	·	·	·	◌	·	10/10
#4 claude-opus-4-8 Anthropic	·	10	·	·	·	·	◌	·	10/10

❯ The Theaterwatch & vote

Same prompt, same clock, side by side. Pick a capability, watch the runs, cast your vote.

❯ ArenaAI judge panel

	Model	ELO	Record
#1	claude-fable-5 Max effort	1065	8W–2L
#2	claude-opus-4-8	1002	3W–3L
#3	claude-fable-5 high effort	1000	2W–2L
#4	claude-opus-5	1000	2W–2L
#5	gpt-5.6 Sol Ultra	998	5W–5L
#6	kimi-code/k3	935	0W–6L

20 matches, blind pairwise, majority of three judges. Every vote is public on the Arena page.

❯ On the benchscoring in progress

claude-fable-5 high effort

Anthropic

4 of 8 runs captured · scoring pending

claude-opus-5

Anthropic

5 of 8 runs captured · scoring pending

qwen3.8-max-preview

Alibaba

awaiting first run

❯ Category breakdownsame points, grouped

The eight frozen tests, grouped by what they measure. Points are the scores, unchanged — no category is scored separately. Every panel runs in season order on the category's full scale.

Game dev

Test 01 — of 10

Creative coding

Tests 02 · 04 — of 20

UI & design

Tests 03 · 05 — of 20

// Pending.

Agentic

Tests 06 · 08 — of 20

Debugging

Test 07 — of 10

// Pending.

❯ Efficiencyrecorded, never scored

Token efficiency

Output tokens per point. Lower is better.

2.2kgpt-5.6 Sol Ultra 34.9kclaude-fable-5 Max effort 18.1kclaude-opus-4-8

Time on the bench

Wall clock across scored tests. Lower is better.

73mgpt-5.6 Sol Ultra 125mclaude-fable-5 Max effort 15mkimi-code/k3 21mclaude-opus-4-8

Model	Points	Time	Output tokens	Tokens / point	Cost
gpt-5.6 Sol Ultra	48	72m 33s	107k	2.2k	·
claude-fable-5 Max effort	34	125m 9s	1186k	34.9k	·
kimi-code/k3	10	14m 55s	·	·	·
claude-opus-4-8	10	20m 47s	181k	18.1k	$4.52