Kimi K3 Gets a 30% Price Hike as Engy’s Demand Outpaces Supply

Engy’s Kimi K3 rates are rising 30% on September 2 at 08:00 UTC because the subnet’s capacity keeps expanding.

Input climbs from $1.50 to $1.95 per million tokens, output from $7.50 to $9.75, cached reads from $0.15 to $0.195.

That single-model hike is Kimi K3’s demand outrunning Engy’s B300 and H200 capacity, the hardware the model needs at scale.

Even after the increase, Kimi K3 on Engy undercuts Moonshot’s official pricing and most aggregator hosts by a comfortable margin.

What’s Driving the Squeeze

• Kimi K3 is Moonshot’s 2.8-trillion-parameter MoE flagship, with a 1M-token context window and native vision support.
• It’s also the model that an independent evaluation from BlueTAO rated best (compared to Chutes) for coding across the entire Bittensor inference stack, scoring 77 out of 77 on Engy.

• Frontier routes like Kimi K3 need B300 and H200 clusters specifically. The rest of Engy’s catalog runs on clustered consumer and mid-range cards, including 4090s, 5090s, and RTX PRO 6000s.
• Engy’s inference proof, described in its public technical documentation, requires every response to carry a Merkle-pinned activation fingerprint, so the network can’t just swap in cheaper hardware to close the gap. That’s why K3 hit the limit.

The Numbers After the Hike

Moonshot’s official card, Engy’s live rate card, and one public benchmark all point in the same direction.

• Moonshot’s official Kimi K3 API charges $3.00 input, $15.00 output, and $0.30 per cache hit, flat across the full 1M window.
• Engy’s post-hike card, $1.95 input and $9.75 output, still lands roughly 35% below that official rate on both ends.
• Kimi K3 through the cheapest OpenRouter listing sat near $2.55 input and $12.75 output as of mid-2026, above Engy’s new numbers.
• Chutes billed the same model at $15.00 output in the public BlueTAO comparison, exactly double what Engy charges after the increase.
• That same BlueTAO benchmark clocked Kimi K3 finishing a four-task coding bundle in 382 seconds on Engy, against 1,176 seconds on Chutes.

• The rest of Engy’s catalog haven’t changed as at this publication, as seen below:

BlueTAO’s builder tested every text model on both Chutes & Engy and picked Kimi K3 on Engy as the top coding option.

More on Engy (SN53):