Moonshot AI Releases Kimi K3.1 Pro, AA Intelligence Index Score of 55 Surpasses GPT-6 Astra
Moonshot AI has released Kimi K3.1 Pro, an upgraded version of Kimi K3, featuring a MoE architecture with 2.8 trillion total parameters and 104 billion activated parameters, activating 16 of 896 experts at a time. According to Artificial Analysis benchmarks, the model ranks fourth on the Intelligence Index with a score of 55, higher than GPT-6 Astra's 53, while the previous generation Kimi K3 scored 44. It ranks first on Terminal Bench 4.0 with 67%, compared with 13% for the previous generation and 64% for Claude Sonnet 5.5; pricing is $2 per million input tokens and $10 per million output tokens, but its output speed of 73 tokens/s is below the average of 81 tokens/s.
Body · AI translation
PC body_June Mouser (in Article)//
AD Manager | AD178089411676 //
// AD Manager | AD178089411676
Chinese artificial intelligence (AI) company Moonshot AI has unveiled Kimi K3.1 Pro, an upgraded version of its Kimi K3 model. Just as its predecessor made a significant impact on the global AI industry, Kimi K3.1 has outperformed OpenAI's GPT-6 Astra in independent evaluations, approaching the performance of Anthropic's Claude Opus 5.5 and Sonnet 5.5.
The new model is an enhanced version of Kimi K3, released in July, developed through post-training. It uses a mixture-of-experts (MoE) architecture with 2.8 trillion total parameters and 104 billion active parameters. Of its 896 experts, 16 are selectively activated during token processing. Moonshot AI said it allocated more computing resources to post-training than for its predecessor while improving training processes and infrastructure efficiency.
According to benchmarks from Artificial Analysis (AA), an independent AI model evaluation organization, Kimi K3.1 Pro ranked fourth with a score of 55 on the Intelligence Index, a composite measure of model performance. The ranking includes Meta Muse Wonder, which has yet to release its API. Kimi K3.1 Pro scored ahead of GPT-6 Astra, which received 53 points, as well as Claude Fable 5.1 and Google's Gemini 4 Argon.
Its predecessor, Kimi K3, scored 44 with reasoning effort set to maximum. Kimi K3.1 Pro also achieved 1,846 Elo on GDPval-AA, which measures performance on practical work tasks, an improvement of 312 Elo over K3.
Moonshot AI has unveiled Kimi K3.1 Pro, which ranks ahead of GPT-6 Astra and Gemini 4 Argon in Artificial Analysis' overall evaluation. (Source: AA)
Kimi K3.1 Pro is priced at $2 per million input tokens and $10 per million output tokens. AA calculated an average cost of $2.11 per task based on its Intelligence Index evaluation, compared with $5.98 for Claude Opus 5.5 at maximum reasoning effort.
However, Kimi K3.1 Pro's output speed of 73 tokens per second is below the average of 81 tokens per second. It also generated 150 million tokens during AA's Intelligence Index evaluation, exceeding the median of 87 million. This suggests that despite competitive per-token pricing, extended reasoning and high output-token consumption could increase actual costs.
"Kimi K3.1 Pro delivers top-tier performance at a reasonable price compared with other models in a similar price range, but its drawbacks include considerably slower speeds and very lengthy responses," AA said.
While Kimi K3.1 Pro performed strongly across most benchmarks, reflecting its high Intelligence Index score, it ranked first in Terminal Bench 4.0. The benchmark evaluates whether AI agents can complete coding tasks in a terminal, or command-line environment, measuring agentic coding and terminal-use capabilities.
Kimi K3.1 Pro scored 67%, up sharply from 13% for its predecessor, Kimi K3, and ahead of Anthropic's Claude Sonnet 5.5 at 64%, giving it the highest score among all models. Moonshot AI also claimed that Kimi K3.1 Pro outperformed both competing models on OfficeQA Pro v2, a benchmark assessing complex financial analysis capabilities.
Kimi K3.1 Pro ranked first in Terminal Bench 4.0, which evaluates agentic coding capabilities in a command-line interface (CLI) environment. (Source: AA)
Moonshot AI also emphasized its adoption of technologies designed to improve hardware resource efficiency. The model incorporates Kimi Delta Attention, the company's implementation of Gated DeltaNet, a memory-saving technology. It reduces memory consumption required for attention computations, improving efficiency when processing long contexts. The technology was also used in Kimi K3.
Kimi K3.1 Pro can also generate animations rather than being limited to static 3D models. According to Moonshot AI's internal tests, the model generated animated simulations of human movements and created several 3D video games.
Meanwhile, the US government, Anthropic and OpenAI have accused China's Moonshot AI of unauthorized distillation of their models. Distillation is a technique that improves another model's performance by using outputs from a more capable model as training data. Anthropic alleged that Moonshot AI used fraudulent accounts and proxy networks to collect responses and reasoning traces from its Claude models and used them to train Kimi.