Kimi K3.1 Pro scores 55 on Artificial Analysis Index, fourth place behind Claude Opus 5.5, with API prices cut by a thir

Moonshot AI has released Kimi K3.1 Pro, a post-trained version of Kimi K3 with significantly more post-training compute. It scores 55 on the Artificial Analysis Intelligence Index, an 11-point jump over K3, placing fourth. Only Claude Opus 5.5 (58), Claude Sonnet 5.5 (56) and Muse Wonder Full (56) rank higher, according to Artificial Analysis. The report comes from The Decoder.

Specifications and pricing

K3.1 Pro keeps the dimensions of K3: 2.8 trillion total parameters, 104 billion active, a one million token context window, and support for text and image input. Moonshot also ships K3.1 Pro Highspeed, which the company claims runs two to three times faster. Both are available through Kimi.ai, Kimi Work, Kimi Code and the Kimi API.

Item	Kimi-K3.1-Pro	Kimi-K3.1-Pro-Highspeed
Input	$2.00 / 1M tokens	$4.00 / 1M tokens
Output	$10.00 / 1M tokens	$20.00 / 1M tokens
Cache input	$0.20 / 1M tokens	$0.20 / 1M tokens

The standard model's prices are down 33 percent from K3 ($3 input, $15 output, $0.30 cache). Artificial Analysis puts the cost per Intelligence Index task at $1.98, versus $2.00 for K3, because K3.1 Pro generates 14 percent more output tokens. The firm says the model sits on the Pareto frontier, and no model with a comparable parameter count scores higher on the index.

Benchmarks

The largest gains are in agentic work. On GDPval-AA, Artificial Analysis measures 1,846 Elo, 312 points above K3, behind Claude Opus 5.5 (1,866) and Muse Wonder Full (1,857).

Moonshot reports the following results, all with reasoning effort set to "max":

Benchmark	Kimi K3.1 Pro	Claude Opus 5.5	GPT-6 Astra
Terminal-Bench 4.0	65.7	66.4	58.2
SWE Marathon v1.1	51.7	46.3	42.5
DeepSWE v1.1	75.5	74.2	74.1
HLE-Full (with tools)	70.0	67.7	57.2
BrowseComp	92.1	90.8	91.5
OfficeQA Pro v2	64.4	62.2	54.4
GPQA-Diamond	94.2	93.9	96.1
BenchCAD (with tools)	92.7	96.2	95.9
ExploitBench	74.7	91.0	100

K3 scored 13 percent on Terminal-Bench 4.0. On Moonshot's internal Kimi Hallu Bench Hard, claim-level hallucinations fell from 23.4 percent with K3 to 8.3 percent, according to the company. Moonshot describes this as a stress test, not a measure of general error rates. GPT-6 Astra scores 5.5 percent on the same test.

On internal finance benchmarks, K3.1 Pro gained 7.2 points on Financial Modeling and 11.6 points on Financial Research, per Moonshot.

3D modeling

Moonshot says K3.1 Pro works directly with Unity, Unreal Engine and Godot, and can write and run Blender and CAD scripts without MCP connectors. On BenchCAD without tools it scores 72.8 percent, against 73.0 percent for Claude Opus 5.5. Moonshot also launched "3D Make" on Kimi.com for previewing, editing and exporting 3D models.

Cybersecurity limits

Moonshot says vulnerability research capability kept growing with post-training scale, because cyber-related content was in the mid-training corpus and related tasks were in the RL environment mix. For the open-weights release, it removed cyber-related data from both training phases. On its internal xdev-bench-hard, the uncleaned model solves 90.5 percent of tasks and the cleaned K3.1 Pro solves 81.0 percent. DeepSeek-V4.1-Flash scores 85.7 percent and Kimi K3 scored 33.3 percent. Moonshot says no open-weights model should push the absolute frontier of offensive cyber capabilities.

Open weights

Moonshot plans to release the weights by November 4, 2026. If it does, Artificial Analysis says K3.1 Pro would be the highest-scoring open-weights model, narrowing the gap to top proprietary models from 12 points to 3, assuming nothing else changes.

What this means

K3.1 Pro delivers a higher Intelligence Index score at roughly the same cost per task as K3, so the 33 percent price cut offsets the extra output tokens. The model's weakest results against Claude Opus 5.5 and GPT-6 Astra are in ExploitBench, BenchCAD with tools and GPQA-Diamond, while it leads on several agentic and office benchmarks Moonshot reports. Most of the headline benchmark figures come from Moonshot, not independent testing. The Intelligence Index and GDPval-AA numbers come from Artificial Analysis. The open-weights claim depends on Moonshot meeting its November 4 target, and the cyber data removal means the released weights are the cleaned version.