Reaction
DeepSeek V4.1 Flash 1017 reportedly scores 9.2% on WeirdML v3
A user says it is well ahead of Kimi-K3 and the strongest Chinese model recorded on the benchmark.
TLDR
A user reported that DeepSeek V4.1 Flash 1017 scored 9.2% on WeirdML v3, well ahead of Kimi-K3. The test used codex-cli and a provider with a limited context window. The user suspected a native harness might improve token efficiency and said they were running a full-context test through PrimeIntellect. Another user wanted to see it beat Haiku, citing Vals cyber results.
Combined views
—
1 Source, first seen ago
