Kimi K3.1 Pro and MiniMax M3.1-Flash Arrive as K2 Retires

Two changes landed in the same week, and they point in opposite directions: Moonshot moved its flagship forward and closed its previous generation, while MiniMax extended its price ladder downward.

Kimi K3.1 Pro

Moonshot describes K3.1 Pro as the first model in a new K3.1 series and its “most capable 3T-class open-weight model to date”, with a 1,048,576-token context window and vision input. It is a post-trained version of Kimi K3 with substantially larger post-training compute, and it ships with a highspeed variant.

Pricing is flat pay-as-you-go with no context-length tiering: ¥13 per 1M input, ¥68 per 1M output, ¥1.3 cached input on the China platform. The highspeed variant lists ¥26 / ¥136 for roughly 2–3x throughput. Batch inference is 60% of the standard price for kimi-k3.1-pro and kimi-k3, and cache writes are billed separately by TTL (5 minutes or 1 hour).

At an indicative 7.1 CNY/USD that is about $1.83 / $9.58 against Kimi K3’s $3.00 / $15.00 — roughly a third cheaper on both sides, which matches Moonshot’s own framing of the release. The international USD list had not been published when this record was created, so the CNY figure is the only verified number and the model page carries it as CNY rather than converting it into a dollar figure the provider has not quoted.

K2 is closing

The same page carries a notice that from 2026-10-31 the K2 series is no longer available to newly registered users, and that all K2 models are scheduled for retirement by the end of December 2026. Existing users are unaffected until then.

For anyone running K2.7 Code or K2.6 in production, that is a migration deadline with roughly eight weeks on it — short enough that the work should be scheduled now, long enough that panic is optional.

MiniMax M3.1-Flash

MiniMax added M3.1-Flash at $0.15 input / $0.60 output per 1M tokens, with prompt-cache reads at $0.015. M3’s tiers are unchanged: $0.30/$1.20 below 512K input, $0.60/$2.40 above, after the permanent 50% discount.

The new model’s announcement also implies a repositioning: the flash tier now sits at the market’s $0.15 floor, while the M3 flagship holds its price. MiniMax is competing on the cheap tier rather than cutting the flagship.

What it means

Two vendors, two strategies. Moonshot is consolidating around a single flagship family and retiring the tier below it; MiniMax is widening its ladder downward. A buyer standardising on one vendor is choosing between concentration and optionality, not just between models.

The K2 retirement is the urgent item. Migration work has a deadline, and the replacement is more expensive: K3.1 Pro is a 3T-class flagship, not a mid-tier coding model. Teams that relied on K2.7 Code’s price will need to re-run their unit economics, not just swap a model ID.

The K3.1 cut is real, and it is smaller than it looks. A third off a flagship that K2 users are being pushed toward still leaves them paying several times the K2.7 Code rate they had budgeted. The migration is mandatory by December; the cost increase is not covered by the discount.