Kimi K3.1 Pro's "parity" is a press release until the weights land

When a lab tells you its model is "just behind" the best in the world, the interesting question isn't how far behind. It's who held the ruler. With Kimi K3.1 Pro, the ruler was mostly Moonshot's own, and that changes how much of the story you should believe.

Moonshot's case rests mostly on Moonshot's own numbers

Moonshot released K3.1 Pro, an upgraded open-weight model, and says full weights will arrive by November 4. On Terminal-Bench 4.0, The Next Web reports, the company says it scored 65.7%, ahead of OpenAI's GPT-6 Astra at 58.2% and just behind Anthropic's Claude Opus 5.5 at 66.4%. The same piece notes those figures are Moonshot's own and haven't been checked independently.

There is one outside data point. According to The Decoder, Artificial Analysis puts K3.1 Pro fourth on its Intelligence Index with 55 points, behind Claude Opus 5.5 (58), Claude Sonnet 5.5 (56) and Muse Wonder Full (56). If the weights ship, the open-versus-proprietary gap on that index shrinks from 12 points to 3.

The gap is narrowing, and that part is real

We don't want to be cynical for sport. An 11-point jump over K3 with the same size, 2.8 trillion total parameters, is a serious result, and it comes from a third party, not a launch slide. Even if you shave a point or two off for generosity, the direction is clear: the distance between what you can download and what you must rent is shrinking fast.

That matters more than any single leaderboard position. It puts pressure on closed labs to justify their premium on something other than raw scores.

But "matches" is doing a lot of work

Look at what's been selected. Moonshot's tests include internal ones: Kimi Hallu Bench Hard, where it reports hallucinations falling from 23.4% to 8.3%, and a vulnerability benchmark of its own. Moonshot itself says the hallucination test is a stress test, not a measure of everyday error rates. And on that same test, GPT-6 Astra still scores better at 5.5%.

Even the friendly coverage hedges. Analytics Vidhya's review concludes K3.1 Pro is "not a clean sweep": the closed models still lead on tool-assisted 3D work, and many of the boldest claims rest on Kimi's internal tests. Its own verdict is that the real test comes when independent teams get the weights. We agree, and we'd go further: a benchmark that the vendor designs, runs and publishes is marketing with decimals.

There's also a quirk worth noticing. Artificial Analysis has K3.1 Pro at 67.2% on Terminal-Bench 4.0, according to OfficeChai, higher than Moonshot's own 65.7%. That's good news for Moonshot, but it also suggests how much these numbers can move depending on setup and who's running them.

The price is the real headline, with an asterisk

The strongest argument for K3.1 Pro isn't parity; it's cost. Moonshot's API prices are $2 per million input tokens and $10 per million output tokens. OfficeChai puts Claude Opus 5.5 at $4/$20 and GPT-6 Astra at $10/$50. That's roughly half the former and a fifth of the latter.

The asterisk: per-token price isn't per-task price. Artificial Analysis found K3.1 Pro uses 14% more output tokens than K3, so a task on its index costs $1.98 versus $2.00 for the old model. A cheaper rate card doesn't automatically mean a cheaper bill, and we'd want to see real workloads before calling it a bargain. The Highspeed variant, at $4/$20, costs the same as Opus 5.5.

Test it on your own workload, not on the launch chart

Our advice is plain. If you run AI at scale, put K3.1 Pro on your shortlist and test it on your own tasks, with your own prompts, measuring cost per completed job. Don't switch on the strength of a launch chart.

Two other caveats deserve weight. The open weights haven't shipped yet, so "open" is still a promise until November 4. And Moonshot is under pressure in Washington: in July, per The Next Web, the White House accused it of using banned Nvidia chips and copying an Anthropic model, without publishing evidence. That is an allegation, not a finding, but any company with compliance obligations will need to weigh it.

Our verdict: the frontier gap is narrowing, and that's the story worth taking seriously. "Parity" is not yet established. Ask again on November 5, when someone other than Moonshot holds the ruler.