🤖 StepFun Releases Step 5 Preview — a Reasoning LLM with 1M Token Context
The Chinese AI studio presented a proprietary 600-billion parameter reasoning model with 'text + image' input on September 18. In the Artificial Analysis Intelligence Index v4.3 benchmark — 44 points, 25th place out of 200 with a median of 25 for the class.
🌍 Prices are aggressive: $1.00 per 1M input tokens and $2.70 per 1M output tokens (medians — $1.88 and $10.00). The cheapening of reasoning models is accelerating, and the key metric becomes the task cost — $0.71 according to AA methodology.
👤 A candidate for comparison for long documents and agentic tasks. Nuances: the model is verbose (approximately 2× the median output tokens), speed is not disclosed, one API provider is available; the evidence base is a single aggregated index without a breakdown by tasks.
Source 1: https://artificialanalysis.ai/models/step-5 Source 2: https://news.ycombinator.com/item?id=49763660