The same practical tasks. Shared places for equal points.
Rounds
20/30
Models
6
Checks
690
Critical
19
20 shared scored tasks
Published standings: position and movement, model and critical misses, checks passed, total points and latest round points. Statistics count checks and critical misses across all models.
#
Model
Checks
Pts / round
1=
DeepSeek V4 Pro0 critical
115/115100% met
115+7
1=
GPT-6 Astra0 critical
115/115100% met
115+7
1=
Grok 4.70 critical
115/115100% met
115+7
4=
Claude Sonnet 50 critical
113/11598% met
113+7
5=
Gemini 4 Argon2 critical
110/11596% met
106+7
6=
Mistral Large17 critical
85/11574% met
51−3
Points = checks met − 2 per critical miss. Signed points: latest round. Arrows: places moved. Check and critical totals cover all models. Shared tasks only; not overall model quality.
Claude, Grok, and Qwen say a recruitment report need not refute the Nvidia board line. Gemini treats the claims as conflicting. None of the six picks a winner from the text alone.
Unscored. Compare reasoning and source use; no winner or League points.
Product names and logos identify their owners. No affiliation or endorsement.
Your AI Playbook
Free member workspace
Keep the setup. Get back to the task.
Your recipes, usual tone and company details in one place. Return to a client email or weekly report with less to fill in.