A $0.14 Model Just Scored 50 — One Point Behind GPT-5.6 Luna
DeepSeek's V4 Flash 0731 scored 50 on the Artificial Analysis Intelligence Index at $0.14/$0.28 per million tokens — one point behind GPT-5.6 Luna and cheaper on both ends.

Moonshot AI's assistant on the K3 model, with 1M-token context and agent swarms
Kimi is Moonshot AI's chatbot, now running the Kimi K3 model released in July 2026, which Moonshot describes as the largest open-weight model with a 1-million-token context window. The app adds deep research, slide generation, document handling, plugins, scheduled tasks and agent swarms.
Latest: Moonshot released the Kimi K3 open-weight model in July 2026.
Third-party evaluations rank K3 near the top; value is held back by a limited free tier.
See all Kimi alternatives, ranked →
DeepSeek's V4 Flash 0731 scored 50 on the Artificial Analysis Intelligence Index at $0.14/$0.28 per million tokens — one point behind GPT-5.6 Luna and cheaper on both ends.
Anthropic released the Claude 4.5 model family between September and November 2025, starting with Claude Sonnet 4.5 on September 29, 2025. This family includes 3 main models: * Claude Sonnet 4.5 (Sept 29, 2025): Announced as a top-tier model for coding and complex agent tasks. * Claude Haiku 4.5
OpenAI dropped GPT-5.2 in December 2025. The update focused on three main areas: reasoning, coding, and long-document handling. It also claimed roughly 30% lower hallucination rates compared to the previous generation. Let’s start with where it actually performs. Strengths of GPT 5.2 1. Strong r