Kimi K3.1 Pro: 5.37/10, $31.88 per run, open weights in November

AICodeKing put Moonshot's Kimi K3.1 Pro through all six KingBench 4 tasks and scored it 5.37 out of 10. That places it sixth out of eleven models on the channel's chart, behind Opus, Astra, Sol, Argon and even Haiku 5.5.

The setup matters for reading the number. The whole benchmark ran through PandaCode, the channel's open-source app for coding agents, tuned for Kimi Code. Six K3.1 workers ran in parallel, with a separate evaluator checking every project.

The launch blog told a different story. It claimed 65.7% on Terminal-Bench 4.0, close to Opus 5.5 at 66.4% and ahead of GPT-6 Astra at 58.2%. The channel says that "almost Opus" framing does not hold up in real projects.

Not everything failed. According to the channel, the model was honest about what it did not build. In the Docker task it blocked all 20 malicious images. In the keyboard app it spoke the real VIA protocol.

The failures are concrete. It built only the Game Boy part of the emulator. Its Docker alternative has no build step and no networking. The Pokémon Red project has broken levelling and a title screen that wipes saves.

Cost is the part that surprises. The full run came to about $31.88. Haiku 5.5 cost $7.40 on the same setup, and DeepSeek V4.1 Flash came in under $2.

Most of that bill is not the code it writes. It is 110 million cached tokens that the model re-reads on every step.

Kimi K3.1 Pro is Moonshot's post-trained K3, with a 1M context and a Highspeed version the channel describes as 2–3× faster. Open weights were promised by November 4th.

According to AICodeKing, once the weights land it will be the best-scoring open-weight model on the chart, ahead of DeepSeek Flash and GLM. That is the channel's projection, not a confirmed result.

The channel's verdict is narrower: a good open model, not a frontier one.

On the API, the channel's advice is blunt: Haiku or DeepSeek Flash are the smarter picks for now. They score better or cost far less, and the gap in price is large.

Kimi K3.1 Pro makes sense if open weights are a hard requirement, or if you want a model that flags its own gaps. It does not make sense as a default for everyday coding at $31.88 a run.

What to check before switching: whether the November weights actually ship, and whether the cached-token cost drops in your own workload. Until then, the launch blog's numbers are not the ones to plan around.