Fetching from the wire…
01
02
03
04
05
06
07
08
Agents2026-06-23 · source-backed
MiniMax-M3's 59.0% edges Kimi K2.6's 58.6%, but those come from vendor-tuned harnesses. On Scale AI's standardized leaderboard with identical scaffolding for every model, the top open-weights entry is qwen3-coder-480b at just 38.7%. (MorphLLM) That 10-to-30-point gap is mostly context-retrieval and tool-use quality in the harness, not raw model capability. This is the single most important caveat to every coding benchmark you'll read this week, including the GLM-5.2 numbers up top. Read scores against the scaffolding that produced them.
Each link below shares sources, entities, or timing with this story.
MiniMax competes with Anthropic / Shared entities / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (MiniMax competes with Anthropic); both cover Bench Pro, Kimi K2, SWE; overlapping topics (benchmark, capability, coding, model).
Ollama supports MiniMax / Shared entities / Shared topic / Earlier coverage
Linked by a graph relationship (Ollama supports MiniMax); both cover Bench Pro, GLM, MiniMax, SWE; overlapping topics (against, benchmark, coding, model).
MiniMax competes with Anthropic / Shared entities / Shared topic / What happened next
Linked by a graph relationship (MiniMax competes with Anthropic); both cover Bench Pro, GLM, SWE; overlapping topics (coding, different, model).
MiniMax competes with Anthropic / Shared entities / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (MiniMax competes with Anthropic); both cover GLM, SWE; overlapping topics (benchmark, coding, model, open-weight).
Moonshot AI deprecates Kimi K2 / Shared entities / Shared topic / Earlier coverage / Tension
Linked by a graph relationship (Moonshot AI deprecates Kimi K2); both cover Bench Pro, GLM, Kimi K2, SWE; overlapping topics (benchmark, coding, model, open-weight).
MiniMax competes with Anthropic / Shared entities / Shared topic / Earlier coverage
Linked by a graph relationship (MiniMax competes with Anthropic); both cover Bench Pro, GLM, SWE; overlapping topics (benchmark, coding, model, open-weight).
MiniMax competes with Anthropic / Shared entities / Shared topic / What happened next
Linked by a graph relationship (MiniMax competes with Anthropic); both cover GLM, SWE; overlapping topics (coding, model).
MiniMax competes with Anthropic / Shared entities / Shared topic / Earlier coverage
Linked by a graph relationship (MiniMax competes with Anthropic); both cover GLM, Kimi K2; overlapping topics (coding, model).