Moonshot Cuts AI Costs by 33% While Closing Gap With Anthropic

The tension in the artificial intelligence market has shifted from pure capability to cost-efficiency. As proprietary leaders like Anthropic hold the top of the leaderboard, Chinese challenger Moonshot AI is leveraging aggressive price cuts to close the performance gap. The company’s latest release, Kimi K3.1 Pro, signals a strategic pivot toward value, offering near-top-tier intelligence at a significantly reduced cost for enterprise users.

Moonshot AI has launched Kimi K3.1 Pro, a post-trained iteration of its previous model that scores 55 points on the Artificial Analysis Intelligence Index. This represents an 11-point improvement over its predecessor, Kimi K3. The score places the model in fourth position, trailing only Claude Opus 5.5 (58 points), Claude Sonnet 5.5 (56 points), and Muse Wonder Full (56 points). For a national audience, this means a Chinese-built model is now within striking distance of the industry’s most expensive and highly regarded competitors.

The financial implications are immediate. While the model retains the same 2.8 trillion total parameter count and 104 billion active parameters as its predecessor, Moonshot has reduced API pricing by 33 percent. Input costs have dropped from $3 to $2 per million tokens, and output costs have fallen from $15 to $10. Cache hits are now priced at $0.20, down from $0.30. These reductions make high-performance AI more accessible for startups and mid-sized businesses that were previously priced out of the premium tier.

Performance gains are most evident in agentic tasks, where the model demonstrates significant maturity. On the GDPval-AA benchmark, Kimi K3.1 Pro achieves an Elo rating of 1,846, a 312-point leap from the previous version. This score sits just behind Claude Opus 5.5 (1,866) and Muse Wonder Full (1,857). In terminal coding tasks, the model reports a 65.7 percent success rate on Terminal-Bench 4.0, surpassing GPT-6 Astra’s 58.2 percent and coming within 0.7 percent of Claude Opus 5.5’s 66.4 percent. By comparison, the older Kimi K3 managed only 13 percent on the same test.

Reliability has also improved, with hallucination rates dropping sharply. According to Moonshot’s internal stress test, claim-level hallucinations fell from 23.4 percent in K3 to 8.3 percent in K3.1 Pro. While GPT-6 Astra still leads with a 5.5 percent rate, the reduction suggests the model is becoming more dependable for business applications. The company notes that this specific benchmark is a stress test rather than a measure of general error rates, but the downward trend is notable for risk-averse enterprise clients.

Beyond text and code, the model has expanded into 3D modeling and direct integration with game engines such as Unity, Unreal Engine, and Godot. On the BenchCAD benchmark, it scores 92.7 percent when using tools, a result that closely follows Claude Opus 5.5 (96.2 percent) and GPT-6 Astra (95.9 percent). Without external tools, it achieves 72.8 percent, nearly matching Claude Opus 5.5’s 73.0 percent. Moonshot claims the model excels at parametric modeling for mechanical parts and architecture, and it has launched a '3D Make' feature on its platform to allow users to preview and export these models.

A key strategic move for Moonshot is the planned release of open-weights by November 4, 2026. If executed, Kimi K3.1 Pro would become the highest-scoring open-weights model available. This would narrow the gap between open and proprietary models from 12 points to just 3 points, assuming no other major releases occur in the interim. This move could pressure proprietary vendors to justify their premium pricing through features other than raw intelligence scores.

The company has also addressed cybersecurity concerns, a growing market anxiety. Moonshot stated that as post-training scaled, vulnerability research capabilities increased naturally. For the open-weights release, the company removed cyber-related data from training to manage offensive potential. On its internal xdev-bench-hard test, the uncleaned version solved 90.5 percent of tasks, while the cleaned K3.1 Pro solves 81.0 percent. This is higher than DeepSeek-V4.1-Flash (85.7 percent) and significantly higher than the previous Kimi K3 (33.3 percent).

The launch includes a 'Highspeed' variant that operates two to three times faster, providing options for latency-sensitive applications. Both versions are available immediately through various Moonshot platforms. For investors and market analysts, the release underscores a trend where Chinese AI firms are competing not just on benchmark scores, but on the total cost of ownership, forcing a re-evaluation of the value proposition for top-tier proprietary models.