Moonshot AI's Kimi K3 Targets Anthropic's Opus 4.8

Moonshot AI has launched Kimi K3, a 2.8-trillion-parameter reasoning model the company says can match or surpass Anthropic's Opus 4.8. The release lands alongside a steep valuation jump for Moonshot, which is now raising capital at $31.5 billion, up from $20 billion just two months ago.

What launched

Kimi K3 went live on July 16, 2026, according to Artificial Analysis, the same day TechCrunch reported the model was expected "in the coming days." It's a reasoning model built on extended thinking / chain-of-thought techniques, and it's multimodal — accepting text and image input while generating text output.

Key specs reported around launch:

Parameters: 2.8 trillion (pre-launch estimates ranged 2–3 trillion)
Context window: 1.0 million tokens
Intelligence Index score: 57
Output speed: 62.0 tokens/second
Time to first token: 1.99 seconds
Pricing: $3.00 per 1M input tokens, $15.00 per 1M output tokens
Evaluation cost: $2,690.80, generating 130M output tokens during testing

Moonshot's prior Kimi K2 models were reportedly well received in the open-source AI community, and the company counts OpenAI, Anthropic, DeepSeek, and Z.ai among its competitors.

Funding trajectory

Moonshot raised $2 billion in May 2026 at a $20 billion valuation. Just two months later, the company is raising fresh capital at $31.5 billion — a jump that coincides with the Kimi K3 launch but whose investor details and intended use of funds aren't disclosed in available reporting.

Sources differ on key details

Two notable conflicts emerge from current reporting:

Open-weight vs. proprietary: TechCrunch describes Kimi K3 as poised to be "the largest open-weight AI model from China," while Artificial Analysis lists it as proprietary, with model weights not publicly available. This is a meaningful discrepancy for any developer planning to build on or fine-tune the model.
Launch timing: TechCrunch's July 16 report frames the launch as imminent ("in the coming days"), while Artificial Analysis lists July 16 as the actual release date — a minor but notable timing mismatch.

No independent benchmark comparison between Kimi K3 and Opus 4.8 is currently available, so the parity claim remains a pre-launch expectation rather than a verified result. Similarly, there's no context for how the Intelligence Index score of 57 stacks up against Opus 4.8 or other competing models.

Why founders should care

For early-stage teams evaluating LLM infrastructure, Kimi K3's specs suggest it could plausibly be a cost-competitive option for document-heavy or long-context reasoning tasks, given its 1.0M-token context window and $3/$15 per-million-token pricing. The chain-of-thought design may also suit agentic or multi-step workflows, though this is unverified until independent benchmarks emerge.

The rapid valuation run-up — from $20B to $31.5B in roughly two months — could indicate growing investor confidence in Chinese open-model ecosystems, which is worth monitoring if you're weighing infrastructure or partnership decisions. But it's also possible the valuation reflects expectations that haven't yet been tested against real-world performance, so founders should treat the Opus 4.8-parity claims with some caution until third-party evaluations are published.

Most critically: given the unresolved open-weight vs. proprietary conflict, founders considering Kimi K3 for products that depend on model weight access — self-hosting, fine-tuning, or on-premise deployment — should verify licensing terms directly with Moonshot before committing engineering resources.

What's still unclear

No direct benchmark data comparing Kimi K3 to Opus 4.8 has been published.
Details of the new funding round — investors, structure, use of funds — remain undisclosed beyond the valuation figure.
The basis for calling Kimi K3 "the largest open-weight AI model from China" is unclear given the proprietary-status conflict.

Founders evaluating Kimi K3 for production use should watch for independent benchmarks and clarified licensing terms before making commitments.