[ ABORT TO HUD ]
SEQ. 1
SEQ. 2
SEQ. 3

Kimi K3 vs Claude: The July 2026 Scoreboard

🇨🇳 China Frontier Models15 min160 BASE XP

The Headline, Stated Precisely

On July 14, 2026, Moonshot AI released Kimi K3 — a 2.8-trillion-parameter open-weight MoE model with a 1M-token context window and native vision. Within days it produced the strongest benchmark results any open-weight model has posted against Claude: K3 beats Claude Fable 5 on 6 of 14 shared benchmarks, tops the Arena.ai Frontend Code Arena at #1 (1,679 points), and edges past Claude Opus 4.8 on the Artificial Analysis Intelligence Index (57 vs 56). Claude Fable 5 still wins the other 8 of 14 head-to-head benchmarks and leads that same index at 60.

Where Kimi K3 Scores Higher Than Claude Fable 5

BenchmarkKimi K3Claude Fable 5Margin
BrowseComp (web research)91.2 (state of the art)88.0+3.2
SWE Marathon (long-horizon agentic coding)42.035.0+7.0
Terminal-Bench 2.188.384.6+3.7
Automation Bench30.829.1+1.7
Program Bench77.876.8+1.0
SpreadsheetBench 234.834.7+0.1

Pattern: K3's wins cluster in agentic execution — long-horizon coding runs, terminal work, browser research, and workflow automation.

Where Kimi K3 Scores Higher Than Claude Opus 4.8

BenchmarkKimi K3Claude Opus 4.8
Artificial Analysis Intelligence Index v4.15756
Terminal-Bench 2.188.384.6
SWE Marathon42.040.0
Online Exp Bench75.565.9
DECK-Bench73.566.9
Finance-Bench62.658.4

Where Claude Fable 5 Still Leads

BenchmarkClaude Fable 5Kimi K3Margin
GDPval-AA v2 (professional work, Elo)1,7601,668+92 — widest gap of any benchmark
FrontierSWE86.681.2+5.4
JobBench57.452.9+4.5
Kimi Code Bench 2.076.972.9+4.0
Zerobench (visual, pass@5)46.041.0+5.0
DeepSWE70.067.5+2.5
CharXiv (visual, with tools)93.591.3+2.2
AA-Briefcase (Elo)1,5831,548+35

Overall standing on the Artificial Analysis Intelligence Index v4.1: Claude Fable 5: 60 · GPT-5.6 Sol: 59 · Kimi K3: 57 · Claude Opus 4.8: 56 · Kimi K2.6: 44. Moonshot's own launch materials place K3 third overall, behind Fable 5 and GPT-5.6 Sol — the wins above are per-benchmark, not an overall crown.

The Economics Are the Real Story

Per 1M tokensKimi K3Claude Fable 5
Input (cache miss)$3.00$10.00
Input (cache hit)$0.30$1.00
Output$15.00$50.00

K3 is roughly 3.3x cheaper across the board at a 1M-token context. On BrowseComp, its state-of-the-art 91.2 comes in at under $2 per task, where competing frontier models run $5–$27 per task.

Known Caveats (Report These Too)

  • Hallucination rate rose: 39% → 51% versus Kimi K2.6 per Artificial Analysis, alongside a documented tendency toward excessive proactiveness in agentic use.
  • Weights not yet public at launch: open weights were announced for late July 2026 — until independent reruns land, all scores are launch-report numbers.
  • Benchmark selection matters: "wins 6 of 14" and "loses the overall index" are both true. Route per task, not per headline.
💡 Operator's Read: For agentic browsing, terminal automation, and long-horizon coding loops, K3 is now a serious Claude alternative at a third of the price. For frontier-difficulty engineering, professional knowledge work, and visual reasoning, Claude Fable 5 still holds the lead. Run both through your own eval gates before switching anything in production.
KNOWLEDGE CHECK
QUERY 1 // 3
On how many of the 14 shared benchmarks did Kimi K3 beat Claude Fable 5 at launch?
All 14
6 of 14
10 of 14
None
Kimi K3 vs Claude: The July 2026 Scoreboard Tutorial | China Frontier Models — Open Source AI Academy