BenchLM's independent snapshot (dated September 4, 2026) catalogs MiniMax M3 as released June 1, 2026, with open-weight status, 1M-token context, and API pricing of $0.30 input / $1.20 output per million tokens ($0.06 cached input, $0.75 blended). The model scores 63.9/100 overall, ranking 49th of 232 tracked models, a The catalog flags that 27 source-displayable benchmark rows are published, with some tracked slots empty and the Reasoning, Math, Multilingual, and Instruction-following categories marked as not eligible or not ranked. Capability percentiles within eligible cohorts are 30th (Agentic), 63rd (Coding), and 65th (Knowledge

MiniMax officially released MiniMax M3 on June 1, 2026, positioning it as a frontier-level model for coding and agentic work. According to the MiniMax research blog, M3 introduces MiniMax Sparse Attention (MSA), a new attention architecture from the MiniMax team that enables context windows of up to 1 million tokens, a The blog details meaningful gains over the prior M2 in coding benchmarks for bug fixing, frontend/backend development, and performance optimization, and reports strong agentic performance on office workflows (search, Office-suite tasks) and initial usability in financial-domain tasks. M3 is available immediately throug