Moonshot AI cuts costs for top-tier model

Moonshot AI has launched Kimi K3.1 Pro, a reasoning model that positions the Chinese company as a serious competitor in the high-end artificial intelligence market. The new release scores 55 points on the Artificial Analysis Intelligence Index, placing it in fourth place globally. This result trails only three other systems, including Anthropic's Claude Opus 5.5, which holds the top spot with 58 points. The model represents an 11-point increase over its predecessor, Kimi K3, signaling a significant leap in capability for the vendor.

For businesses evaluating API costs, the pricing structure offers a notable adjustment. Moonshot has reduced token prices by 33 percent across the board. Input costs now stand at $2 per million tokens, down from $3. Output costs have dropped to $10 from $15, while cache hits are priced at $0.20, a decrease from $0.30. Despite these lower unit costs, the total expense for a standard Intelligence Index task remains nearly unchanged at $1.98, up slightly from $2.00 for the previous version. This marginal increase is attributed to the model generating 14 percent more output tokens during inference.

The company maintains the same architectural footprint as the prior iteration, utilizing 2.8 trillion total parameters with 104 billion active ones. The context window remains at one million tokens, supporting both text and image inputs. According to Artificial Analysis, this configuration places the model on the Pareto frontier, meaning no other system with a comparable parameter count achieves a higher intelligence score. A high-speed variant, named K3.1 Pro Highspeed, is also available and reportedly runs two to three times faster than the standard version.

Performance gains are most evident in agentic tasks and coding environments. On the GDPval-AA benchmark, the model achieves an Elo score of 1,846, which is 312 points higher than its predecessor. It now sits just behind Claude Opus 5.5 and Muse Wonder Full. In terminal-based coding tests, the system reports a success rate of 65.7 percent, surpassing GPT-6 Astra and approaching the performance of the leading proprietary models. Hallucination rates on internal stress tests have also decreased significantly, falling from 23.4 percent to 8.3 percent.

The model demonstrates strong capabilities in 3D modeling and finance. It can interact directly with game engines such as Unity and Unreal Engine for project builds. On the BenchCAD benchmark, it scores 92.7 percent when using tools, a figure close to top competitors. Moonshot also notes improvements in financial modeling and research tasks. The company plans to release the open model weights by November 4, 2026. If no other major updates occur, this release would make Kimi K3.1 Pro the highest-scoring open-weights model, narrowing the gap between open and proprietary systems from 12 points to just three.

Moonshot has also addressed cybersecurity concerns in its technical documentation. The company removed cyber-related data from training phases for the open-weights version to manage offensive capabilities. On internal benchmarks reproducing recent software vulnerabilities, the cleaned model solves 81.0 percent of tasks, compared to 90.5 percent for the uncleaned version. This strategy aims to balance defensive utility with a cautious approach to offensive potential, a trend that mirrors discussions in the broader AI industry regarding the power of open models.