GLM 4.7 and MiniMax M2.1 launch
Merry Christmas and happy New Year! We're pleased to launch support for some great new models: GLM-4.7 and MiniMax M2.1. They're very strong updates to already-good open-source model lines, and represent a new frontier for coding agents. We're also launching a new, more flexible way to pay for usage.
GLM 4.7
GLM-4.7 is one of the strongest models for coding agents. It matches or beats Claude 4.5 and GPT-5.1 on most major benchmarks related to math and coding, and it's a massive leap forward from their previous model, GLM-4.6.
And aside from being technically stronger, it also produces much more beautiful output; for example, see the difference between GLM-4.7's take on a 3D voxel garden, vs GLM-4.6:
GLM 4.7 Kyoto Voxel Garden:
GLM 4.6 Kyoto Voxel Garden:
While both are technically-correct voxels, I think you'll agree that GLM-4.7's version is by far more beautiful. Benchmarks alone don't quite capture the difference.
MiniMax M2.1
MiniMax M2.1 is a strong coding model with especially good benchmarks for coding iOS and Android apps. It approaches Claude 4.5 and DeepSeek 3.2 on web and terminal benchmarks, and is significantly stronger at coding for iOS and Android. It's also just a much more intelligent model than their previous release, M2. It's a 229B-parameter model, but since only 10B of those parameters are active at any given time, it's very quick.
Flexible usage billing
By popular demand from our Discord, we're launching a new, flexible option for using both subscriptions and usage-based billing. We're adding a new toggle to the usage-based billing portal:
This new toggle allows you to fall back to usage-based, per-token billing once you exhaust your susbcription's rate limits, instead of getting HTTP 429 rate limit errors. We don't think this is the right choice for everyone: it means that you have less-predictable monthly costs, since instead of simply stopping when you run out of your rate limit, instead you get charged more. And if you're regularly overshooting, we think it's much more affordable to just upgrade to a higher-tier subscription. But for some use cases this makes a lot of sense! If you already bought credits and want to use them, or if you rarely exceed rate limits but do so occasionally, this can help.
If you got this far, thanks for reading! We've got a lot of new features planned for the New Year, and we're excited to show them to you. See you next time,
♥️ Matt