Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
Official1
minimax.ioofficialauthority2,159 words
MiniMax M2.7 is the first model in the M2-series to deeply participate in its own self-evolution by building complex agent harnesses and completing elaborate productivity tasks. On the SWE-Pro benchmark, M2.7 scored 56.22%, nearly approaching Opus's best level, while achieving an ELO score of 1495 on GDPval-AA, the highest among open-source models. The model maintains a 97% skill adherence rate while working with over 40 complex skills, each exceeding 2,000 tokens, demonstrating significant improvements in complex editing for the Office suite.written, reviewed, cited
- M2.7 autonomously executed an iterative optimization loop for over 100 rounds to discover effective programming performance improvements.
- Using M2.7 for live production debugging reduced the recovery time for system incidents to under three minutes on multiple occasions.
Analysis9
firethering.com2026-04-12HN 84 pts · 36 comments1,101 words
MiniMax released M2.7, an agentic model that improved its own code over 100 rounds without human direction, achieving a 30% performance boost. The model won 9 gold medals, 5 silver medals, and 1 bronze across 22 machine learning competitions runnable on a single A30 GPU. It scores 56.22% on SWE-Pro and reduces live production incident recovery time to under three minutes on multiple occasions. Commercial use requires prior written authorization from MiniMax, while non-commercial use remains free.written, reviewed, cited
- M2.7 scores 76.5 on SWE Multilingual and 52.7 on Multi SWE Bench, testing closer to real world engineering scenarios.
- The model maintains 97% skill compliance across 40 complex skills on MM Claw, each skill exceeding 2,000 tokens.
andlukyane.com2026-05-20HN 30 pts · 2 comments2,224 words
MiniMax M2.7 excels in agentic workflows when constraints are explicit and output formats are concrete, but it struggles when important context is left implicit. The model ran on MiniMax's Plus tier ($40/month) and processed roughly 91M total tokens at PAYG rates of $0.30/$1.20 per million input/output. It returned tool calls noticeably faster than Opus 4.7, subjectively around 2x.written, reviewed, cited
- The model assumes exposed test rows are the entire set and hardcodes them without explicit prompt constraints.
- A 100-line prompt tuned for Opus 4.7 does not transfer one-to-one to M2.7 without a bootstrap tuning step.
- The model inferred a domain tag rule for F1 scores even though such a rule does not exist in the taxonomy.
openrouter.ai2026-03-18authorityHN 7 pts · 1 comments437 words
MiniMax M2.7 is a next-generation large language model designed for autonomous, real-world productivity and continuous improvement. It features a 205K context window and was released on Mar 18, 2026, with input and output pricing of $0.21 / $0.84 per 1M tokens. The model achieves a 1495 ELO on GDPval-AA and delivers 56.2% on SWE-Pro while handling workflows like live debugging and financial modeling.written, reviewed, cited
inferencex.semianalysis.comauthority2,143 words
MiniMax released M2.5 and M2.7, sparse mixture-of-experts models with 229.9B total parameters and 9.8B activated per token. M2.5 launched on 2026-02-12 under a modified MIT license, while M2.7 followed on 2026-03-18 with a NON-COMMERCIAL LICENSE requiring attribution for commercial use. Both models target agentic and coding tasks, achieving 80.2% on SWE-Bench Verified.written, reviewed, cited
- M2.7's license prohibits commercial use without prior written authorization from MiniMax.
- The M2.7 model card reports an internal version autonomously optimizing a programming scaffold over 100+ rounds.
- M2.5-Lightning sustains 100 tokens/second at $0.30/M input and $2.40/M output.
artificialanalysis.aiauthority972 words
codejam.info2026-04-18HN 6 pts · 2 comments1,457 words
GPT-5.4 achieved an average cleanliness score of 9 while GLM-5.1 scored 8 in the technical approach evaluation. Kimi K2.5 and MiniMax M2.7 were consistently ranked at the bottom with average scores of 5.6 and 5.2 respectively.generated, cited
artificialanalysis.ai2026-03-25authority780 words
snackonai.com2026-05-273,708 words
chatforest.com2026-05-134,670 words
