Qwen3.5 122B-A10B vs Kimi K2 Thinking vs MiniMax-M2.5-highspeed

Qwen3.5 122B-A10B is the better all-round choice, scoring 58/100 against Kimi K2 Thinking (42) and MiniMax-M2.5-highspeed (41). It leads on inputs & features. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.

MiniMax-M2.5-highspeed is cheaper at $0.60 input / $2.40 output per million tokens (official MiniMax (minimax.io) API price). Kimi K2 Thinking costs $0.60 input / $2.50 output per million tokens (median across 10 API providers); Qwen3.5 122B-A10B costs $0.40 input / $3.20 output per million tokens (official Alibaba API price). At a typical mix of three input tokens to one output token, that is $1.05 per million tokens for MiniMax-M2.5-highspeed versus $1.07 for Kimi K2 Thinking (1× as much) and $1.10 for Qwen3.5 122B-A10B (1× as much).

There is no independent benchmark that covers all three models yet. Qwen3.5 122B-A10B has not been scored yet, Kimi K2 Thinking has an ECI of 146.0 and MiniMax-M2.5-highspeed has not been scored yet.

There are no published SWE-bench Verified results for Qwen3.5 122B-A10B, Kimi K2 Thinking and MiniMax-M2.5-highspeed yet, so there is no like-for-like coding score. Test both on a sample of your own tasks. All three support tool calling for agent workflows.

Qwen3.5 122B-A10B and Kimi K2 Thinking have the largest context windows (262,144 and 262,144 tokens), against 204,800 for MiniMax-M2.5-highspeed. Maximum output per response: Qwen3.5 122B-A10B up to 65,536, Kimi K2 Thinking up to 262,144, MiniMax-M2.5-highspeed up to 131,072 tokens.

Qwen3.5 122B-A10B accepts text, images, audio and video; Kimi K2 Thinking accepts text; MiniMax-M2.5-highspeed accepts text. Qwen3.5 122B-A10B handles the widest range of inputs.

Yes, all three publish their weights, so you can self-host them.

Qwen3.5 122B-A10B is the newest, released Feb 23, 2026. MiniMax-M2.5-highspeed came out Feb 13, 2026; Kimi K2 Thinking came out Nov 6, 2025. Knowledge cutoff: Kimi K2 Thinking Aug 2024.

Each model gets a 0–100 score on capability (50%, independent benchmark results); price (25%, blended price per million tokens (3 input : 1 output), log scale); inputs & features (15%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (10%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Oct 4, 2026.