Moonshot Releases Kimi K3.1 Pro, Open-Weights Model Nipping at Claude Flagships

On October 24, Moonshot released Kimi K3.1 Pro, positioning it as "the strongest 3T-class open-weight model to date." Built on a MoE architecture with 2.8T total parameters and 104B activated, it supports 1M token context and multimodality, with a focus on whole-project coding "from idea to finished product" (Unreal/Unity/Godot games, Blender, CAD, etc.). Weights are scheduled to open on November 4.

Confirmed

• Benchmarks: Official charts show K3.1 Pro topping SWE Marathon at 51.7, ahead of Opus 5.5 (46.3) and Fable 5.1 (45.6); it beats Opus 5.5 on 9 of 14 benchmarks and GPT-6 Astra on 10 (relayed by haider1). Terminal-Bench 4.0 comes in at 65.7%, beating GPT-6 Astra (58.2%) and trailing Opus 5.5 by only 0.7 points, at roughly $4–5.5 per task—about a third of Claude's cost.
• Specialized skills: OfficeQA Pro v2 at 64.4 for SOTA, beating Opus 5.5 (62.2); financial modeling up 7.2 points and financial research up 11.6 points over K3; BenchCAD at 92.7%, close to top closed-source models, with direct operation of Blender, Unreal, Godot, CAD, and other tools.
• Pricing: API at $2.00/MTok input and $10.00/MTok output—two-thirds of K3's price—with cache hits at $0.20/MTok; also launched the 2–3x faster K3.1 Pro Highspeed ($4.00/MTok input). Artificial Analysis confirmed an across-the-board 33% price cut versus K3, with per-task cost at $1.98, on par with the previous generation.
• Technical details: The official blog describes K3.1 Pro as a post-trained version of K3, with substantially more post-training compute, joint large-scale RL across hundreds of downstream domains, and improvements to the RL framework, long-context reasoning, training efficiency, and sandbox infrastructure (also relayed by natolambert).

Unconfirmed

• No independent external evaluations yet; results mainly come from official benchmark charts. Despite the price cut, bindureddy argues it falls short of a 5x reduction and DeepSeek still wins on value.

Why it matters

• If independently verified, K3.1 Pro becomes another open-weight benchmark against closed-source flagships (Opus 5.5, GPT-6 Astra), delivering near-parity performance at roughly 1/3–1/4 the cost. The announcement also moved related prediction markets on Polymarket. Community testing after weights open on November 4 will be the key validation point.

2026-10-24 ~ 2026-10-24 · 30 related posts

Full story(7 episodes)→

• Episode 1: Kimi K3.1 leak frenzy intensifies as official teaser and internal test screenshots surface(2026-10-22, 16 posts)
• Episode 2: Moonshot teases Kimi K3.1 after strong Arena showing(2026-10-23, 4 posts)
• Episode 3: Kimi K3.1 Pro Rolls Out: 3T-Parameter Flagship with 1M Context(2026-10-24, 12 posts)
• Episode 4: Moonshot Releases Kimi K3.1 Pro, Open-Weights Model Nipping at Claude Flagships(2026-10-24, 30 posts)
• Episode 5: Nathan Lambert Calls Kimi K3.1 Pro the Strongest Open-Weight Model(2026-10-24, 2 posts)
• Episode 6: Kimi K3.1 Pro Ranks 4th on AA Intelligence Index with Major Agent Gains(2026-10-24, 9 posts)
• Episode 7: Artificial Analysis Publishes Full Kimi K3.1 Pro Benchmark Charts(2026-10-24, 2 posts)

• Kimi 3.1 priced at $2/$10 per 1M tokens, still not competitive vs DeepSeek, says investor — bindureddy · 2026-10-24
• Kimi K3.1 Pro Benchmarks: Tops SWE Marathon at 51.7, Near Opus 5.5 Across Coding Boards — scaling01 · 2026-10-24
• Kimi K3.1 Pro launches with 1M context, scores 65.7% on Terminal-Bench 4.0, beats GPT-6 Astra — cedric_chee · 2026-10-24
• Moonshot Launches Kimi K3.1 Pro: 2.8T-Param Open-Weight Model Rivals GPT-6 and Opus 5.5 — Kimi_Moonshot · 2026-10-24
• Kimi K3.1 Pro Scores 65.7% on Terminal-Bench 4.0 at Roughly a Quarter of Opus 5.5's Cost — Kimi_Moonshot · 2026-10-24
• Kimi K3.1 Pro Hits 92.7% on BenchCAD, Working Natively in Blender, Unreal and CAD — Kimi_Moonshot · 2026-10-24
• Kimi K3.1 Pro Claims OfficeQA Pro v2 SOTA With 64.4, Boosting Finance Tasks Over K3 — Kimi_Moonshot · 2026-10-24
• [source] Kimi K3.1 Pro API Pricing: $2/MToK Input, Two-Thirds of K3's Price — Kimi_Moonshot · 2026-10-24
• [source] Kimi Publishes K3.1 Pro Tech Blog Detailing Large-Scale Joint RL Training Stack — Kimi_Moonshot · 2026-10-24
• Kimi K3.1 Pro benchmark sweep: tops SWE Marathon, HLE and 5 more, called best open-weight model — inductionheads · 2026-10-24
• Moonshot releases Kimi K3.1 Pro: tops SWE Marathon, pushes open intelligence — Recoil42 · 2026-10-24
• Moonshot AI releases Kimi K3.1 Pro, positioning it as open intelligence — Recoil42 · 2026-10-24
• Kimi K3.1 Pro tech blog: stack-wide RL scaling, 65.7% on Terminal-Bench 4.0 — natolambert · 2026-10-24
• Kimi K3.1 Pro debuts: beats Claude Opus 5.5 on Terminal-Bench 4.0 at ~1/3 the cost — Yuchenj_UW · 2026-10-24
• Moonshot's Kimi K3.1 Pro: 2.8T params, 1M context, priced 50% below Claude Opus 5.5 — kimmonismus · 2026-10-24
• Kimi K3.1 Pro tops Terminal-Bench 4.0 at roughly a third of Claude's cost — Pyros-SD-Models · 2026-10-24
• Kimi K3.1 Pro debuts with 3T params, matching frontier models on agentic coding — karminski3 · 2026-10-24
• Kimi K3.1 Pro leaks: 65.7 on Terminal-Bench 4.0 near Opus 5.5 at a fraction of the cost — realmrfakename · 2026-10-24
• Kimi K3.1 Pro open weights due Nov 4; claimed to beat Opus 5.5 on 9/14 benchmarks — haider1 · 2026-10-24
• Kimi K3.1 Pro tops SWE Marathon and agent benchmarks, fueling open-source 'last time they catch up' debate — cephaloform · 2026-10-24
• Kimi K3.1 Pro cuts token prices 33%, cost per task holds at $1.98 — ArtificialAnlys · 2026-10-24
• Kimi K3.1 Pro fully rolled out with Highspeed tier; open weights due November 4 — cedric_chee · 2026-10-24
• Moonshot releases Kimi K3.1 Pro, claims coding benchmark win over GPT-6 Astra — Polymarket · 2026-10-24
• Moonshot's open-weight Kimi K3.1 Pro hits 65.7% on Terminal-Bench, a point behind Opus at 1/3 cost — Prompt Engineering · 2026-10-24

6 near-duplicate retellings: Kimi_Moonshot · Scobleizer · No-Head-Royal · mark_k · karminski3 · BenBajarin