MiniMax M3 (MiniMax)

LLM Gateway's own blog post on OpenCode Go pricing confirms that MiniMax M3 is part of the active model catalog on the platform, grouped alongside MiniMax M2.7, Kimi K2.7, Qwen3.7 Max, GLM-5.2 and others in the $60-per-month allocation tier. The post explains that the $10/month Go plan now meters usage by dollar at eac Because metering runs at per-token rates, the article stresses that how far the $60 ceiling stretches depends entirely on which model is picked, framing M3 alongside higher-cost entries like Grok 4.5 and GPT-5.6 Luna. It also notes that past a limit, requests fall back to free models (currently Ox Alpha Free, marked li

The Towards AI piece dated June 4, 2026 frames MiniMax M3's release inside a broader news roundup alongside OpenAI Codex expansion and Microsoft's MAI-Thinking-1. It reports that MiniMax positioned M3 with "frontier coding and agentic scores" and a 1-million-token context window, and treats the release as one that mate The article's emphasis is on which releases change the daily loop for builders, which benchmarks are actually being compared, and which cost structure makes agentic work more usable. In that framing, M3 is grouped with reasoning-style releases aimed at long-context agentic coding workflows, with its 1M context claim ci

BenchLM's model page for MiniMax M3 (data as of September 22, 2026) reports a composite capability score of 55.3/100, ranking the model 60th of 196 tracked models, with pricing of $0.30 per million input tokens, $1.20 per million output tokens, and $0.06 per million cached tokens (blended $0.75). It cites a measured th The tracker breaks out verified category rankings: Agentic 39.8 (rank 45 of 105, 58th percentile, 9 verified benchmarks), Coding 39.6 (rank 63 of 135, 54th percentile, 10 verified benchmarks), Multimodal 52.0 (rank 36 of 50, 29th percentile, 5 verified benchmarks), and Knowledge 48.0 (rank 62 of 160, 62nd percentile, 2

evals.report's model page for MiniMax M3 tracks 10 reported benchmark scores with explicit source-status flags, listing SWE-bench Pro at 59.0% (Verified, Jun 1, 2026), Terminal-Bench 2.1 at 66.0% (Verified), MCP Atlas at 74.2% (Verified), AA-Omniscience Index 1 (Official), Artificial Analysis Intelligence Index 55 (Off As an independent tracker, the page adds measurable post-launch signal beyond the official announcement: it distinguishes verified reproductions (SWE-bench Pro, Terminal-Bench 2.1, MCP Atlas, Design Arena) from official-but-untested figures and from weaker or unverified entries such as OSWorld 70.1% and DeepSWE 13.3%.

MiniMax's official Investor Relations news index serves as a first-party record tying MiniMax M3 to specific benchmark results. The Jun 9, 2026 MaxProof blog post explicitly states: "In the M3 release post, we reported the performance of the M3 model on two international mathematical olympiad benchmarks: IMO 2025 and U The IR page also documents the broader MiniMax model timeline, with the Aug 26, 2026 first-half financial results announcement, the Aug 13, 2026 MiniMax Music 3.0 open-weights music model launch, and the Jul 30 / Aug 3, 2026 MiniMax H3 omni-modal video model release — separate model products in the MiniMax family rathe