Breaking

Chinese Startup Moonshot AI Releases Kimi K3, Rattling Bets on Anthropic's Trillion-Dollar Valuation

July 17, 2026 at 19:04 EDT

Foundation Models
Open Source
Funding & M&A

Moonshot AI, a Chinese startup, released its flagship Kimi K3 model on July 16, 2026, a launch that has cast fresh doubt on the sky-high valuations of leading American AI companies such as Anthropic. The debut prompted prediction-market traders to trim the perceived odds that Anthropic reaches a $1.5 trillion valuation by year-end, with the probability sliding to roughly 67% after having climbed as high as 78% in earlier trading.

July 16, 2026 · Moonshot AI

Kimi K3 Puts a Chinese Model at the Frontier — and Cracks in U.S. AI Valuations

Moonshot AI's 2.8-trillion-parameter open-weight model lands third overall on independent leaderboards at a fraction of the price — enough to move prediction markets on whether Anthropic hits a $1.5T valuation.

2.8T

parameters, mixture-of-experts

token context window, flat pricing

overall on Artificial Analysis

6.3×

faster decoding at long context

The price gap that moved the market

Output cost per million tokens — Kimi K3 lists at roughly a third to a half of comparable U.S. offerings.

$15

Kimi K3

output / 1M

~$30–45

Comparable U.S. models

2–3× the cost

Input tokens listed at $3 / 1M · flat across the full 1M-token context.

Prediction market: odds Anthropic hits $1.5T by year-end

78%

Earlier peak

→

67%

After K3 launch

Anthropic raised in May 2026 at a $965B post-money valuation.

Why developers are excited

Tops the Frontend Code Arena; 2nd on Vals Index
Strong coding & agentic tasks — 3D and multiplayer prototypes
Open weights due July 27 for customization

The caveats

Pricey by Chinese standards vs GLM-5.2
Less mature surrounding tool ecosystem
Open weights raise misuse & jailbreak concerns

What to watch

The outlook hinges on the July 27 open-weight release and whether benchmark strength turns into durable real-world adoption against entrenched U.S. incumbents. Bears see margin compression on U.S. models; bulls argue cheap frontier models could expand total compute demand.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.