Released: 2026-03-03
google/gemini-3-flash-preview
2
4
9
7
1
3
8
6
5
8
3
5
2
4
6
7
9
1
1
7
6
8
5
9
4
2
3
5
9
2
6
3
7
1
8
4
3
8
4
9
2
1
6
7
5
6
1
7
4
8
5
9
3
2
7
2
3
5
6
8
5
1
9
4
6
8
1
7
2
3
5
9
9
5
1
3
9
4
6
7
2
google/gemini-2.0-flash-001
2
4
9
7
1
5
8
6
3
3
7
5
2
4
6
1
9
8
1
8
6
3
9
7
4
2
5
6
9
4
5
3
2
7
8
1
8
3
2
6
7
1
4
5
9
5
1
7
4
8
9
6
3
2
4
6
3
9
2
8
5
1
7
9
2
8
1
7
3
5
4
6
7
5
1
8
6
4
9
3
2
anthropic/claude-3.7-sonnet
anthropic
5
4
9
7
1
3
8
6
2
8
3
2
2
4
6
5
1
7
1
7
6
5
8
9
4
2
3
6
9
4
5
3
2
1
8
7
2
8
3
6
4
1
7
9
5
4
1
7
8
5
7
9
3
6
9
6
3
1
2
8
2
5
4
3
5
8
4
7
9
3
2
1
7
2
1
3
9
4
6
7
2
openai/gpt-4o
openai
3
5
9
7
4
1
8
6
2
8
4
7
2
9
6
5
3
1
1
6
2
5
8
3
4
9
7
4
9
6
1
3
5
2
8
7
2
3
8
6
7
1
3
4
5
5
1
7
8
2
9
9
7
6
7
2
3
4
5
8
1
2
9
6
8
5
3
7
2
7
1
4
9
7
1
9
6
4
6
5
3
mistralai/mistral-medium-3
mistral
5
4
9
7
2
3
8
6
1
8
3
7
2
9
6
5
1
4
1
2
6
8
5
4
4
9
7
4
9
2
6
3
5
7
8
1
6
8
3
4
7
1
2
9
5
3
5
7
9
8
2
6
4
8
9
6
3
5
4
8
1
2
7
2
1
8
3
7
9
4
5
6
7
4
1
6
1
4
6
3
2
meta-llama/llama-3.3-70b-instruct
meta
5
8
9
7
4
3
8
6
1
4
3
6
2
8
6
5
9
7
1
5
2
9
6
8
4
3
9
8
9
4
5
3
2
1
8
6
3
1
5
8
9
1
7
4
5
6
2
7
4
1
5
3
1
8
9
4
3
1
5
7
2
5
4
2
6
8
3
7
9
6
1
3
7
5
1
6
2
4
9
7
2
deepseek/deepseek-chat-v3-0324
deepseek
4
3
9
7
5
8
1
6
2
8
5
1
2
9
6
3
4
7
1
7
2
4
6
3
8
5
9
6
9
5
1
3
7
2
8
4
3
4
8
9
2
1
7
6
5
2
1
7
5
8
4
9
3
6
7
6
3
8
1
2
5
9
4
5
2
8
6
7
9
4
1
3
9
8
1
3
4
5
6
7
2
x-ai/grok-3-mini-beta
x-ai
4
5
9
7
1
3
8
6
2
7
1
8
2
4
6
5
3
9
1
6
2
8
9
5
4
7
3
6
9
4
1
3
2
7
8
5
2
8
7
4
5
9
1
3
6
3
4
1
5
8
7
9
2
6
5
7
3
9
2
8
6
4
1
8
3
5
6
7
1
2
9
4
9
2
6
3
5
4
1
8
7
Each model receives the same Sudoku puzzle with a partially filled grid. Models are tasked with solving the puzzle by filling in all empty cells following Sudoku rules, and their solutions are evaluated without feedback or retries.