Moonshot AI releases 2.8T parameter Kimi K3, pricing at $3/$15 per million tokens
Chinese AI lab Moonshot AI released Kimi K3, a 2.8 trillion parameter model priced at $3 per million input tokens and $15 per million output tokens. The model is currently available via API, with open weights promised by July 27, 2026. This represents the most expensive pricing from a Chinese AI lab to date, matching Anthropic's Claude Sonnet series.
Kimi K3 — Quick Specs
Moonshot AI releases 2.8T parameter Kimi K3, pricing at $3/$15 per million tokens
Chinese AI lab Moonshot AI released Kimi K3, a 2.8 trillion parameter model priced at $3 per million input tokens and $15 per million output tokens. The model is currently available via API, with open weights promised by July 27, 2026.
Model specifications and pricing
Kimi K3 represents a significant scale increase from Moonshot's previous model, K2.6, which had 1 trillion parameters and was priced at $0.95 input/$4 output per million tokens. Moonshot claims K3 is their "most capable model to date" and describes it as the first "open 3T-class model," surpassing DeepSeek's 1.6T parameter v4 Pro.
The $3/$15 pricing matches Anthropic's Claude Sonnet series and makes K3 the most expensive model released by a Chinese AI lab. This represents more than a 3x price increase over K2.6.
Benchmark performance
According to Moonshot's self-reported benchmarks, K3 beats Claude Opus 4.8 max and GPT-5.5 high on most tasks, while trailing Claude Fable 5 and GPT-5.6 Sol. Third-party evaluator Artificial Analysis reports K3 achieved an Elo of 1547 on their private long-horizon knowledge work evaluation, placing it behind only Claude Fable 5. The model now leads Arena.ai's Frontend Code arena, surpassing Claude Fable 5.
Artificial Analysis reports K3's cost per task at $0.94, similar to GPT-5.6 Sol ($1.04), roughly half the price of Opus 4.8 ($1.80), though higher than open weight alternatives. The model uses 21% fewer output tokens than K2.6 according to the Artificial Analysis Intelligence Index.
Reasoning token usage
K3 currently offers only one reasoning effort level ("max"), which shows significant token consumption. A simple SVG generation prompt consumed 13,241 reasoning tokens to output 3,417 tokens of response, costing approximately $0.25 for that single request. The same prompt counted as 95 input tokens despite being only 10 words, suggesting an 85-token hidden system prompt.
The model includes vision capabilities, demonstrating strong performance on image description tasks.
What this means
Kimi K3's pricing strategy marks a shift for Chinese AI labs toward premium pricing tiers, matching Western competitors like Anthropic. The aggressive pricing suggests confidence in the model's capabilities, though the high reasoning token consumption at the "max" effort level could make practical applications expensive. The promised open weights release by July 27, 2026, will be significant given the model's 2.8T parameter size—making it the largest openly available model if released as planned. The model's strong performance on coding benchmarks and third-party evaluations suggests Chinese labs continue closing the gap with frontier Western models, though head-to-head comparisons remain challenging without standardized benchmarks.
Related Articles
Alibaba Unveils Qwen3.8-Max, a 2.4T-Parameter Open-Weight Model for Coding and Agentic Work
Alibaba announced Qwen3.8-Max, a 2.4T-parameter flagship model targeting coding and long-horizon agentic work, with open weights promised for next week alongside Qwen3.8-27B. The model posted strong third-party benchmark results, ranking #4 in Frontend Code Arena and matching Claude Opus 4.7 on the Vals Index at roughly 2.3x lower cost.
OpenAI Halts Parts of Astra Model Development After It Hit 'Critical' Cybersecurity Threshold
OpenAI disclosed that its in-development Astra model showed cyberattack capabilities strong enough that it cannot rule out a 'Critical' risk classification. The company has paused related internal activity and added security controls under its Preparedness Framework.
Mistral's 3B-Parameter Shieldstral Matches 20B Safety Model on Text Benchmarks
Mistral's new Shieldstral, a 3-billion-parameter open-weight safety classifier, posts an 84.9% F1 score on text benchmarks—tying OpenAI's GPT-OSS-Safeguard-20B, a model roughly seven times larger. The model lets operators define safety rules at runtime using plain-language yes/no questions instead of fixed taxonomies.
Mistral AI Releases Shieldstral-1.0-3B, a 3B-Parameter Policy-Adaptive Safety Classifier
Mistral AI has released Shieldstral-1.0-3B, a compact open-weight safety classifier that evaluates text and images against natural-language policies specified at inference time. The 3B model runs on a single GPU and reports F1 scores competitive with or exceeding larger moderation models like LlamaGuard-4-12B and GPT-OSS-Safeguard-20B on multiple benchmarks.
Comments
Loading...