I wonder if these are starting to get reasonable enough to use locally?
MiniMax M2.5 released: 80.2% in SWE-bench Verified
31–40 of 60 posts
Re: MiniMax M2.5 released: 80.2% in SWE-bench Verified
#32This is cool, but they mentioned affordability, and said this is about $1/hour to run, which is about what I pay for claude code on $200/mo plan. This is not literally true, sometimes I'm running up to 3 concurrent intermittently throughout the day for maybe 60 hours per week.
So I do believe if there is something that comes up that is literally continuous, would be interesting, but I'm not sure about it right now. I would be curious if anyone has anything they would literally use running 24/7.
Re: MiniMax M2.5 released: 80.2% in SWE-bench Verified
#33Everyone is using this sort of let me group the plots weirdly instead of sorting them to make harder to compare. I see you folks
Re: MiniMax M2.5 released: 80.2% in SWE-bench Verified
#34M2 was one of the most benchmaxxed models we've seen. Huge gap between SWE-B results and tasks it hasn't been trained on. We'll put 2.5 on the list. https://brokk.ai/power-ranking
Re: MiniMax M2.5 released: 80.2% in SWE-bench Verified
#35> M2.5-Lightning [...] costs $0.3 per million input tokens and $2.4 per million output tokens. M2.5 [...] costs half that. Both model versions support caching. Based on output price, the cost of M2.5 is one-tenth to one-twentieth that of Opus, Gemini 3 Pro, and GPT-5. Huge - if not groundbreaking - if the benchmark stats are true.
yes it's good. But you should also look at GLM 5 and Kimi K2.5 when looking at M2.5. It's amazing we have so many good and cheap open weight models now which are really not far behind the top models from the big US AI companies.
Anthropic Claude Code and OpenAI Codex plans are subsidised.
The Chinese open weight models hosted in US or Europe make more sense to use when you want to stay model agnostic and less dependent on a single AI company with relative expensive APIs.
Re: MiniMax M2.5 released: 80.2% in SWE-bench Verified
#36$1/hr sounds suspiciously close to a price of one A100 80GB GPU.
Maybe an 8x node assuming batching >= 8 users per node.
Re: MiniMax M2.5 released: 80.2% in SWE-bench Verified
#37Pelican is recognizable but not great, bicycle frame is missing a bar: https://gist.github.com/simonw/61b7953f29a0b7fee1f232f6d9826...
You should switch to an octopus riding a bike, much harder.
Not an SVG, but I'm pretty impressed by what Gemini 3.0 Fast does: https://gemini.google.com/share/52c1229bd1d9
/imagine an svg of an octopus riding a bike. 1 arm shading its eyes from the sun, another waving a cute white flag, 2 driving the bike, 2 peddling the wheels, and 2 drifting behind in the wind
Re: MiniMax M2.5 released: 80.2% in SWE-bench Verified
#38Earlier quoted context omitted.
That's what I found with some of these LLM models as well. For example I still like to test those models with algorithm problems, and sometimes when they can't actually solve the problem, they will start to hardcode the test cases into the algorithm itself.. Even DeepSeek was doing this at some point, and some of the most recent ones still do this.
Sounds exactly what a junior-dev would do without proper guidance. Could better direction in the prompts help? I find I frequently have to tell it where to put what fixes. IME they make a lot of spaghetti (LLMs and juniors)
wtf kinda juniors are you interacting with
Re: MiniMax M2.5 released: 80.2% in SWE-bench Verified
#39> M2.5-Lightning [...] costs $0.3 per million input tokens and $2.4 per million output tokens. M2.5 [...] costs half that. Both model versions support caching. Based on output price, the cost of M2.5 is one-tenth to one-twentieth that of Opus, Gemini 3 Pro, and GPT-5. Huge - if not groundbreaking - if the benchmark stats are true.
Cost per token doesn't really matter anymore, cost per task it more important.
Re: MiniMax M2.5 released: 80.2% in SWE-bench Verified
#40Earlier quoted context omitted.
Sounds exactly what a junior-dev would do without proper guidance. Could better direction in the prompts help? I find I frequently have to tell it where to put what fixes. IME they make a lot of spaghetti (LLMs and juniors)
wtf kinda juniors are you interacting with
Lots of self-taught; looking for an entry level.