Moonshot AI's Latest Release Could Slash the Gap Between Open and Proprietary AI

Moonshot AI, a Chinese AI company, has released a new reasoning model, Kimi K3.1 Pro, which has immediately secured the fourth position on the Artificial Analysis Intelligence Index with a score of 55 points. This development raises questions about the future landscape of open-weights AI, as the company has stated its intention to release the model's weights by November 4, 2026. Should that plan proceed as scheduled, K3.1 Pro would become the highest-scoring open-weights model available, a move that would shrink the current gap between open and proprietary models from 12 points to just 3.

The model is a post-trained version of its predecessor, Kimi K3, and its capabilities have seen significant enhancement. Its score on the Artificial Analysis Intelligence Index represents an 11-point jump from K3. Currently, only three models rank higher: Claude Opus 5.5, Claude Sonnet 5.5, and Muse Wonder Full. Moonshot also announced a variant, K3.1 Pro Highspeed, which the company claims operates two to three times faster. Both versions are available through Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.

The most pronounced improvements are in agentic tasks. On the GDPval-AA benchmark, K3.1 Pro achieves an Elo of 1,846, a 312-point increase over K3, placing it just behind Claude Opus 5.5 and Muse Wonder Full. On Terminal-Bench 4.0, the model scores 65.7 percent, a figure that is ahead of GPT-6 Astra and narrowly behind Claude Opus 5.5. This is a considerable leap from K3's 13 percent on the same test. The model's performance on tasks involving coding, financial modeling, and 3D modeling has also been noted for substantial gains.

While performance metrics have improved, the model's dimensions remain unchanged. K3.1 Pro retains K3's 2.8 trillion total parameters and 104 billion active parameters, with a context window of one million tokens. The cost, however, has decreased significantly. API prices have been reduced by 33 percent, with input tokens now costing $2 per million, down from $3, and output tokens at $10, down from $15. The company asserts that no model with a comparable parameter count achieves a higher score on the Intelligence Index.

The release also includes specific modifications for cybersecurity. For the open-weights version, Moonshot has stripped cyber-related data from both training phases, a decision that reduces its capability on an internal test that reproduces recently disclosed vulnerabilities. The uncleaned version of the model solves 90.5 percent of tasks on the xdev-bench-hard test, while the cleaned version for release solves 81.0 percent. Moonshot states that no open-weights model should advance the absolute frontier of offensive cyber capabilities, but it also notes that such models serve defensive purposes, such as auditing legacy systems. Anthropic has previously shown that even a model with a lower score can write exploits at near-Claude Mythos levels.