LLM API Pricing Report: 2026-10-08

Twenty models are free to use via OpenRouter as of 2026-10-08, unchanged from the previous day. The list spans providers including Google, NVIDIA, Thinking Machines, Poolside, inclusionAI, Apodex, Dots Studio, Cohere, LiquidAI, and OpenRouter's own Free Models Router. Context windows range from 65,536 tokens for LiquidAI: LFM2.5-2.6B (free) to 1,048,576 tokens for Google's Lyria 3 Pro Preview, Lyria 3 Clip Preview, and Thinking Machines' Inkling Small and Inkling models. NVIDIA contributes five free models, including Nemotron 3.5 Lightning and Nemotron 3 Ultra at 1,000,000 tokens context. No models became free or stopped being free since 2026-10-07.

Anthropic added two variants of Claude Haiku 5.5. The standard version prices input at $0.10 per million tokens and output at $0.50 per million tokens with a 1,000,000 token context window. A batch variant offers input at $0.05 per million tokens and output at $0.25 per million tokens with the same context length. Neither model has open weights. One model, Kwaipilot: KAT-Coder-Pro V2.5, was removed from the catalog.

Eleven models recorded price movements between 2026-10-07 and 2026-10-08. Qwen: Qwen3.6 27B reduced input from $0.32 to $0.30 per million tokens and output from $3.25 to $2.00 per million tokens. Qwen: Qwen3.5-35B-A3B increased input from $0.08 to $0.15 and output from $0.75 to $1.00 per million tokens. Thinking Machines: Inkling raised input from $0.95 to $1.00 per million tokens while output held at $4.05. DeepSeek: DeepSeek Pro Latest moved input from $0.2514 to $0.2909 per million tokens with output steady at $4.20. DeepSeek: DeepSeek Flash Latest dropped input from $0.044 to $0.0152 but raised output from $0.30 to $0.80 per million tokens. Z.ai: GLM 5.3 cut input from $0.07 to $0.062 and output from $7.00 to $2.60 per million tokens. Z.ai: GLM Flash Latest increased input from $0.036 to $0.040 with output unchanged at $0.50. DeepSeek: DeepSeek V3.2 lowered input from $0.28 to $0.259 per million tokens, output unchanged at $0.42. DeepSeek: DeepSeek V4 Flash 0731 raised input from $0.0163 to $0.0180, output steady at $1.28. MoonshotAI: Kimi Latest reduced input from $0.78 to $0.49 per million tokens, output unchanged at $13.00. MoonshotAI: Kimi K3 cut input from $0.79 to $0.50, output steady at $15.00. Z.ai: GLM 5.1 decreased input from $1.40 to $0.966 and output from $4.40 to $3.036 per million tokens.

The lowest input prices among paid models start at $0.0152 per million tokens for DeepSeek: DeepSeek Flash Latest (output $0.80, 1,048,576 token context). IBM: Granite 4.0 Micro follows at $0.017 input and $0.112 output with 131,000 token context and open weights. OpenAI: gpt-oss-20b offers $0.018 input and $0.09 output at 131,072 tokens with open weights. DeepSeek: DeepSeek V4 Flash Latest and DeepSeek: DeepSeek V4 Flash 0731 both sit at $0.018 input and $1.28 output with 1,048,576 token context; the latter has open weights. Mistral: Mistral Nemo prices input at $0.019 and output at $0.03 with 131,072 token context and open weights. inclusionAI: Ling 3.0 Flash VL and Ling 3.0 Flash each charge $0.021 input with outputs of $0.0616 and $0.063 respectively, both at 262,144 token context and open weights.

Five models share the maximum context window of 2,000,000 tokens: OpenRouter's Auto Router (Beta), Pareto Code Router, and Auto Router, plus SpaceXAI: Grok 4.20 Multi-Agent and SpaceXAI: Grok 4.20. These routers and models lead the catalog for context capacity.