LLM API Pricing Report — 2026-10-22
OpenRouter tracks 468 models as of 2026-10-22, with 21 models currently free and 178 offering open weights; one new free model was added, one batch model was removed, and 12 models saw price changes including significant increases for several DeepSeek and Z.ai models.
Free model landscape
The free tier holds 21 models from providers including Google, NVIDIA, Thinking Machines, MiniMax, Dots Studio, Recursive, XingChen, Poolside, Cohere, and OpenRouter itself. Context windows range from 128,000 tokens for NVIDIA Nemotron 3.5 Content Safety up to 1,048,576 tokens for Google Lyria 3 Pro Preview, Google Lyria 3 Clip Preview, Thinking Machines Inkling Small, Thinking Machines Inkling, and several NVIDIA Nemotron 3.5 variants. MiniMax M3.1 Flash offers 1,000,000 tokens. Mid-range contexts of 262,144 tokens are provided by Recursive Flint, XingChen Xing4.0 29B A4B, Poolside Laguna S 2.1 and Laguna XS 2.1, Google Gemma 4 26B A4B and Gemma 4 31B, and NVIDIA Nemotron 3 Super. Cohere North Mini Code and NVIDIA Nemotron 3 Nano Omni sit at 256,000 tokens, while the Free Models Router offers 200,000 tokens.
New and removed models
Recursive Flint (free) entered the catalog with 262,144 token context and open weights. Google Gemini 2.5 Flash Lite (batch) was removed from the tracked list.
Price changes
Twelve models updated pricing since 2026-10-21. Qwen Qwen3 Next 80B A3B Instruct reduced input from $0.10 to $0.09 per million tokens; output held at $1.10. Xiaomi MiMo-V2.5-Pro cut input from $0.435 to $0.3045 and output from $0.87 to $0.609. Z.ai GLM 5.3 Flash lowered input from $0.0471 to $0.0392 and output from $0.12 to $0.0773. Z.ai GLM Flash Latest matched those new rates. Qwen Qwen3.5-9B dropped input from $0.10 to $0.08 and output from $0.15 to $0.13.
DeepSeek Pro Latest raised input from $0.1307 to $0.2235 with output steady at $8.00. Z.ai GLM 5.3 increased input from $0.079 to $0.70 and output from $1.40 to $2.20. MoonshotAI Kimi Latest kept input at $0.238 but lifted output from $16.00 to $24.00. Z.ai GLM 5.1 moved input from $0.966 to $1.40 and output from $3.036 to $4.40. NVIDIA Nemotron 3 Super jumped from free to $0.30 input and $2.40 output. DeepSeek V4.1 Flash rose from $0.056 to $0.07 input and $0.12 to $0.50 output. DeepSeek V4 Flash Latest climbed from $0.0168 to $0.0475 input and $0.1101 to $0.1602 output.
Cheapest paid models by input price
The lowest input prices among paid models start with IBM Granite 4.0 Micro at $0.017 per million tokens input and $0.112 output, 131,000 token context, open weights. OpenAI gpt-oss-20b follows at $0.018 input, $0.09 output, 131,072 context, open weights. InclusionAI Ling 3.0 Flash sits at $0.021 input, $0.063 output, 262,144 context, open weights. OpenAI gpt-oss-20b (batch) is $0.024 input, $0.112 output, same context and openness. Nex AGI Nex-N2.5-Mini asks $0.025 input, $0.10 output, 262,144 context, open weights. OpenAI GPT-5 Nano (batch) prices input at $0.025, output at $0.20, 400,000 context, closed weights. Meta Llama 3.2 1B Instruct comes in at $0.027 input, $0.201 output, 60,000 context, open weights. Mistral Mistral Nemo rounds the top eight at $0.029 input, $0.03 output, 131,072 context, open weights.
Largest context windows
Five models share the maximum 2,000,000 token context: OpenRouter Auto Router (Beta), OpenRouter Pareto Code Router, SpaceXAI Grok 4.20 Multi-Agent, SpaceXAI Grok 4.20, and OpenRouter Auto Router.
Key takeaways
- 21 models are free today, spanning context windows from 128,000 to 1,048,576 tokens.
- Recursive Flint joined the free tier; Google Gemini 2.5 Flash Lite (batch) was removed.
- Six models reduced prices, while six increased, with the largest hikes on DeepSeek Pro Latest, Z.ai GLM 5.3, MoonshotAI Kimi Latest, Z.ai GLM 5.1, NVIDIA Nemotron 3 Super, and DeepSeek V4 variants.
- IBM Granite 4.0 Micro remains the cheapest paid input at $0.017 per million tokens.
- Five models offer a 2,000,000 token context window, all routers or Grok variants.