MiniMax

[red arrow]

https://docs.openclaw.ai/providers/minimax

[craggy plover]

Bumping this up

[stiff frost]

I've been surprisingly impressed with this model on their $20 coding plan. Also spawns its own agents. Just get the identity and soul nailed.

[chrome oxide]

For my needs the starter plan ($10) as primary had been good for me. When there is confusion or misinterpretation it usually comes down to my instructions not being tight.

[stiff frost]

Agreed, I have not hit a rate limit once and busy coding all day.. Weird question, but it seems like my agents are competing for "orchestration" or head agent, whatever I configure. Anyone experiencing this?

[lilac solstice]

Is minimax really 95% as good as opus at 5% of the cost? Seems too good to be true.

[stiff frost]

It's really good with coding and extremely cheap on coding plan. Personality is not greats, so you just need to be Very clear on instructions. I usually let Gemini write a blueprint then feed it to Minimax. That seems to be doing the trick.

[haughty fern]

Minimax canged the coding plan yesterday... not good...

New Model m2.7 and its not good anymore... Before it feelt Limitless. now ist very limited....

[chrome oxide]

What changes are you seeing? It looks like m2.5 is still available. https://platform.minimax.io/docs/guides/models-intro#models-overview
It seems usage and pricing is the same, at least for the starter plan that I'm using (or maybe it was $9, now it's $10)?

[carmine beacon]

I don't see any bad changes to the coding plan, if anything it got better, the usage limits are 3x what they were before.

[ruby delta]

Please join us tonight (in ~6.5 hours) for the AI Ecosystem Sessions with researchers and engineers from Minimax, to discuss their new model Minimax M2.7. Please note, the original announcement had the wrong time listed for Eastern Standard Time. For those on the east-coast, it will be 10pm (EST). We really hope to see you there!

https://discord.gg/clawd?event=1483671881210138685

[idle perch]

@winter yacht Are there plans to make the harness you guys created for the RL stuff available for subscribers?

[finite gyro]

claw eval just post MinMax 2.7 test results https://github.com/explaindio/ClawEval

[potent jewel]

Thanks for posting this. Based on their recommendation I increased my maxTokens setting.

[chrome oxide]

where are we supposed to put maxTokens in openclaw.config - is it under models -> providers -> minimax-portal -> models [] .maxTokens, or is it agents -> defaults -> models -> model -> params .maxToken?

[potent jewel]

The first one - under models. See also here https://docs.openclaw.ai/providers/minimax#minimax-m2-7-api-key

[worn bane]

What thinking level do you recommend with minimax 2.7?

[knotty basin]

hi is minimax 2.7 avaibale in openclaw ?

it just shows 2.5 to me

[soft tangle]

#releases message

[twin oracle]

I was running minimax M2.7 via ollama just fine. But all of a sudden I got an error says M2.7 is no longer connected. Anyone experienced the same ?

[shadow siren]

Hi, I'm using Minimax M2.7-highspeed token plan($40/month) for browser automation and it is very unstable. I mean, I made an MD file for the automation but it ignores some essential commands. I tried to use gpt-5.3-codex through OpenAI OAuth and it worked quite well, but the weekly limit in the Plus version ($20/month) was very low. I also tried to use other models like Kimi v2.5 and openrouter free models but they were even worse than MiniMax. How can I fix it? I'm trying to fix it for 2 weeks and especially trying to change the model for a week but I wasn't able to go through. Is there any approach to make MiniMax work better? My md file is about 13KB and I think it's lower than the limit

[haughty fern]

minimax 2.5 works best for me (Coding plan). but you must setup a lot of rules for all models. search for agent drift...
I tried minimax 2.7 but i don´t like it. wierd things happend.

[chrome oxide]

I noticed this in the latest (2026.3.28) release notes - does this mean I can't continue to use 2.5? Haven't played around with 2.7 yet, but heard some complaints about it... using the minimax coding plan (or whatever they're calling it now) with 2.5 as my primary model. **
MiniMax: trim model catalog to M2.7 only, removing legacy M2, M2.1, M2.5, and VL-01 models. (#54487) Thanks @liyuan97.**
https://github.com/openclaw/openclaw/releases/tag/v2026.3.28

[soft tangle]

I believe you can still use it if you add it manually, its just not there by default anymore. I noticed that too, not sure why all of those were removed

[chrome oxide]

I haven't tried/upgraded yet to the latest, but checking in with krill in #1459642797895319552 it seem to indicate as you said, it should still work if configured (as long as the backend is still serving that version).

[soft tangle]

yeah thats where my answer came from too, except i was talking to my own claw the other day 😅

[shadow siren]

Hi I tried to test the browser automation task with both Minimax M2.7 and GLM-5-Turbo, of course GLM is better than Minimax but it burns 2% of the Max Plan's 5-hours limit - which is approximately 32 prompts, where as Minimax occupies only 72 tool calls (about 4.8 prompts) . Why does GLM spend much more prompts than Minimax in the same task? Is there any approach to reduce it?
P.S. GLM says a lot of messages during the task and how can I reduce the messages if it's good for saving prompts?

[red swallow]

Does anyone use the MiniMax 2.7 100 TPS plan for $40? Have some questions about it.

[red swallow]

How's the speed?

[shadow siren]

It's not 2x faster than the $20 plan. Seems odd

[rapid bough]

It doesn't really feel faster, let's put it that way, but the quality, I don't feel the best, is that there are constant Chinese artifacts in the output.

[dire cliff]

I don't understand how are you guys using this model in openclaw

it's literally saying it does something but it doesn't

[chrome oxide]

@dire cliff Are you using 2.5 or 2.7? And are you using their coding (token) plan, via minimax directly? I'm still on 2.5, the "1500 model requests / 5 hours" plan, and as a primary driver it works well, but that's more for "kick off cron job x" or "why didn't xyz run" or the occassional "run some skill", so maybe I'm not taxing it. Honestly though most of the things I'm doing are with local model qwen35-35b-a3b via cron jobs, with specific instructions. I was using 2.5 to do these originally though... I should kick them off again with 2.5 and 2.7 and see how they do.

[chrome oxide]

I wonder if it would be worth putting a proxy between openclaw and minimax to monitor the traffic and see where it's going off the rails.

[haughty fern]

anyone have problems with minimax coding? the servers are under high load... lot of timeouts

[dire cliff]

I was on the 2.7 from the coding plan yes but nah that’s not for me. Its not doing anything.

[rigid cipher]

Is it not good? Or what's bad about the plan?