Moonshot AI

Tier A · Strong with clear operating rules.

Kimi K3.1 Pro

A-tier coding workhorse with seven inspectable visual-benchmark artifacts across classic briefs and the new Superbench 2.1 contracts.

Try the active visual tests

Open the generated scene to test it, or compare that prompt with another model.

7 runs

Concourse Divide

Run date: 2026-10-26

Generation provenance: unrecorded

Superbench 2.1 brief · tool-free single-message authoring, no repair turn

Visible agent rerouting, collision avoidance, and recovery from blocked connectivity.

Ember Glider

Run date: 2026-10-25

Generation provenance: unrecorded

Superbench 2.1 returning brief · original 1.x run protocol

Sunset gliding journey testing flight energy management, checkpoint flow, and atmospheric scene craft.

Lightkeeper's Atelier

Run date: 2026-10-26

Generation provenance: unrecorded

Superbench 2.1 brief · tool-free single-message authoring, no repair turn

Material differentiation, dynamic light, geometric shadows, and art direction.

Low-Poly Tower Defense

Run date: 2026-10-25

Generation provenance: unrecorded

Superbench 2.1 returning brief · original 1.x run protocol

Diorama tower defense testing economy balance, wave design, placement rules, and combat readability.

Neon Drift

Run date: 2026-10-25

Generation provenance: unrecorded

Synthwave time-trial racing testing drift physics, lap timing, ghost replay, and unlock progression.

Wrecklight Salvage

Run date: 2026-10-26

Generation provenance: unrecorded

Superbench 2.1 brief · tool-free single-message authoring, no repair turn

Three-phase game progression with persistent physical cargo and a visible late-stage consequence.

Archived model versions

The latest Kimi K3.1 Pro runs are shown above. Earlier versions are preserved separately and are not counted in this current run set.

Editorial verdict

Where this model fits

Back to all benchmarks →

Kimi K3.1 Pro replaces K3 in the current editorial roster. Its published evidence includes three classic builds recovered from Kimi Code sessions and four new Superbench 2.1 artifacts made by isolated, tool-free, single-message generation agents. A tier is an editorial routing choice, not a numeric benchmark score.

Best for

  • Long-horizon coding in Kimi Code
  • Interactive HTML, canvas, and WebGL prototypes
  • Single-brief builds with explicit acceptance checks
  • Visual iteration followed by independent review

Watch out

  • The three classic artifacts and four 2.1 artifacts used different delivery protocols
  • Concourse Divide failed the required South-only all-48 recovery; Lightkeeper's Atelier failed WebGL linking and is not a working 3D scene; no aggregate score is assigned
  • Artifact publication is evidence for inspection, not proof of production reliability
  • Confirm current Kimi Code access, pricing, and usage limits before adoption

Why it is ranked here

  1. Seven Kimi K3.1 Pro visual-benchmark artifacts are published: Neon Drift, Ember Glider, Low-Poly Tower Defense, and all four new Superbench 2.1 briefs, including the disclosed Atelier failure.
  2. Each 2.1 artifact came from a fresh agent whose sole creative brief was the frozen canonical prompt. Generation used no tools or reference implementations, and the returned HTML was packaged without code edits or a repair turn.
  3. The A-tier placement reflects the October editorial assessment and the inspectable run set. It does not assign a pass, aggregate quality score, or universal ranking to the model.

Evidence

2026-11-10

Superbash editorial model ranking

Takeaway: Kimi K3.1 Pro is currently placed in Tier A.

The November 2026 editorial roster places Kimi K3.1 Pro at rank 9.

Open source →
2026-10-26

Superbash visual benchmark suite: Kimi K3.1 Pro

Takeaway: Seven benchmark artifacts are published, including one for every new Superbench 2.1 brief and explicit failed-run evidence.

The run record separates the recovered core sessions from the isolated tool-free 2.1 generations and keeps behavioral review distinct from artifact publication.

Open source →

Local benchmark run

Seven visual-benchmark artifacts across classic briefs and new Superbench 2.1 contracts

Superbash Learn sourceOctober 26, 2026Run record and published artifacts →
Sources, pricing details and methodology

Run identity

Model specifications

Codex platform ↗
Status
Available in this Codex session
Codex model ID
k3.1-pro
Run date
October 26, 2026
Access route
Kimi Code desktop. The October 25–26, 2026 benchmark sessions record model k3.1-pro (alias kimi-code/k3.1-pro), including four published Superbench 2.1 artifacts from isolated tool-free generation runs; fresh retries and validation outcomes are disclosed in the run record.

Benchmark sources

Kimi K3.1 Pro has three classic-brief artifacts recovered from saved Kimi Code sessions and four new Superbench 2.1 artifacts, including disclosed Concourse recovery and Atelier WebGL-link failures. Ember Glider and Low-Poly Tower Defense are returning 2.1 briefs under the original protocol; each new 2.1 contract was assigned to a fresh tool-free subagent that received only the canonical prompt and returned one HTML response, published without code edits or a repair turn.

Kimi K3.1 Pro visual benchmark run record

These are inspectable local artifacts, not provider benchmark scores, passed rubrics, or an aggregate rating. The run record separates the two delivery protocols. Per-artifact reports distinguish standalone and browser checks, observed failures, and open items; they are not complete independent rubric review.

Evaluation settings

    Edition coverage

    Recorded means a published artifact, not a behavioral pass. Historical returning results retain their original protocol.

    Superbench 2.1: 6/6 recorded · 0 pending · 0 not run
    • 15 Ember Glider · repository-run · Recorded
    • 14 Low-Poly Tower Defense · repository-run · Recorded
    • 21 Skyhook Cargo · single-prompt · Recorded
    • 22 Concourse Divide · single-prompt · Recorded
    • 23 Lightkeeper's Atelier · single-prompt · Recorded
    • 24 Wrecklight Salvage · single-prompt · Recorded