› Before paying $20 for a subscription, consider two cheaper options: free AI chat plans and low-cost model APIs.
› Using my usage logs from October 24, 2026, I calculated the cost of a task on DeepSeek V4.1 Flash, GLM-5.3, Kimi K3.1 Pro and Qwen3.8 Max, then compared the results with the free versions of ChatGPT, Claude, Gemini and DeepSeek.At a glance (as of October 24, 2026)For $1.27 in API usage, you can get about 1,180 typical short tasks on DeepSeek V4.1 Flash, about 800 on GLM-5.3 and about 170 on Kimi K3.1 Pro.Quality: DeepSeek V4.1 Flash scored 94.8% in my test, compared with 98.7% for Claude Opus 5.5.DeepSeek's free chat uses the same V4.1-Flash-1017 model, so you don't need the API for a conversation.Free plans from major providers offer weaker models: ChatGPT Free includes GPT-5.6 Luna, while Gemini without a subscription mainly uses Flash-Lite.What you get with free AI plansFree planModelProvider's stated limitNotesChatGPT FreeGPT-5.6 Luna"Unlimited" standard chats, subject to abuse safeguards27,000-token context; limited files, images and voiceClaude FreeSonnet and HaikuFive-hour session; no message count specifiedNo Opus or Claude Code; up to five projectsGemini without a subscriptionAuto mode, mainly Flash-LiteStandard limit, refreshed every five hoursMore difficult prompts may be routed to Flash or ProDeepSeekV4.1-Flash-1017No published limitsNo paid chat planKimi AdagioKimiMessage count not specifiedPaid plans start at $19Mistral FreeMistral models"Limited messages and searches"Pro offers up to six times more messagesI've covered the free versions in more detail in separate guides: ChatGPT for free, Gemini for free after the October 9 changes and Claude Haiku 5.5 for free.How much does a task cost on low-cost models?These figures come from my test across 10 languages on October 24, 2026, using OpenRouter with medium reasoning; Qwen and GLM were tested in seven languages.
› Prices are from official price lists on the same date.
› For DeepSeek, I use peak-hour rates ($0.30 / $1.20); off-peak rates, including weekends, are half as much.Model (input/output USD per 1 million tokens)Typical short taskDocument of about 30,000 tokensTasks for $1.27DeepSeek V4.1 Flash ($0.30 / $1.20)about $0.00107about $0.00866about 1,180GLM-5.3 ($1.40 / $4.40)about $0.00157about $0.03856about 806Gemini 4 Argon ($2 / $10)about $0.00429about $0.05387about 297Kimi K3.1 Pro ($2 / $10)about $0.00732about $0.06149about 174Qwen3.8 Max ($2 / $6)about $0.00820about $0.05471about 155In practical terms, $1.27 divided by the $0.0011 cost of a DeepSeek task comes to about 1,180 tasks.