Been almost a month since the last update, now that the dust of August and September's LLM insanity is cleared out, new models are settling in nicely.

A number of updates over last week or so:

• (Boosted) DeepSeek V4.1 Flash switched from timed 4x usage to permanent $60 usage
• (New) MiMo-V2.6 Pro and MiMo-v2.6 Flash were both added with identical limits to V2.5
• (New) Claude Haiku 5.5 was just added but it's more expensive compared to its direct competitor GPT-6 Luna (both $15 credit)

MiMo-V2.6 in benchmarks is a massive leap from MiMo-V2.5 for both Pro and Flash, and it's definitely better no denying it, but I noticed a significant disconnect between real world performance (coding) vs what the benchmarks portray. Sadly MiMo-V2.6 appears to be blatantly bench-maxed (too much) that expectations don't match reality immediately.

Currently for me I like DeepSeek V4.1 Flash for most soft skills like writing automation and researching ideas and it also runs my hermes agent's and in my opinion has the best personality and is blazingly fast (often above 200 t/s!).

GLM-5.3-Flash is my go to for coding, although sometimes Qwen3.8 Flash actually performs better or on par even though its benchmarks put it slightly below, I still perceive GLM-5.3 LLMs as ace for coding.