OpenRouter API Pricing Update: September 30, 2026
On September 30, 2026, OpenRouter tracked 464 models with 20 free models and 178 open-weight models, while introducing four new paid models from OpenAI and adjusting prices for 12 existing models.
Free Models
OpenRouter listed 20 free models as of September 30, 2026, including Google's Lyria 3 Pro Preview and Lyria 3 Clip Preview, both with 1,048,576 context tokens, and Thinking Machines' Inkling Small and Inkling, each with 1,048,576 context tokens.
Other free models include NVIDIA's Nemotron 3.5 Lightning and Nemotron 3 Ultra (1,000,000 context tokens), Dots Studio's Dots3-Note Preview (512,000 context tokens), and multiple models from Qwen, Poolside, Google, NVIDIA, Cohere, and LiquidAI with context tokens ranging from 65,536 to 262,144.
New Models
Four new models were added on September 30, 2026, all from OpenAI under the GPT-6.1 Sol series. The standard GPT-6.1 Sol and GPT-6.1 Sol Pro are priced at $2.00 per million input tokens and $10.00 per million output tokens, each with 1,050,000 context tokens.
Batch versions of these models were also introduced: GPT-6.1 Sol Pro (batch) and GPT-6.1 Sol (batch) both cost $1.00 per million input tokens and $5.00 per million output tokens, maintaining the same 1,050,000 context token limit. None of the new models are open-weight.
Price Changes
Twelve models had price adjustments between September 29 and September 30, 2026. Notable decreases include Qwen: Qwen3 30B A3B Instruct 2507, which dropped input pricing from $0.10 to $0.0481 and output from $0.30 to $0.193 per million tokens, and OpenAI's gpt-oss-120b, which fell from $0.15 to $0.037 input and $0.60 to $0.17 output per million tokens.
Some models saw increases: DeepSeek V3.2 Exp rose from $0.1344 to $0.27 input and $0.2016 to $0.41 output per million tokens, while DeepSeek V4 Flash Latest increased output pricing sharply from $0.32 to $1.25 per million tokens despite a small input drop from $0.018 to $0.012. Z.ai: GLM 5.2 reduced input from $0.6496 to $0.1449 but increased output from $2.0416 to $3.99 per million tokens.
Cheapest Paid Options
The lowest input-cost paid model is DeepSeek: DeepSeek V4 Flash Latest at $0.012 per million input tokens and $1.25 per million output tokens, with 1,310,720 context tokens and non-open weights.
Other low-cost paid models include IBM's Granite 4.0 Micro ($0.017 input, $0.112 output, 131,000 context tokens, open weights), OpenAI's gpt-oss-20b ($0.018 input, $0.09 output, 131,072 context tokens, open weights), and Mistral's Mistral Nemo ($0.019 input, $0.03 output, 131,072 context tokens, open weights).
Key takeaways
- OpenRouter hosted 464 models on September 30, 2026, with 20 available for free and 178 being open-weight.
- Four new GPT-6.1 Sol variants from OpenAI launched at $2.00/$10.00 and $1.00/$5.00 per million input/output tokens for standard and batch versions.
- Twelve models experienced price changes, with the largest input reduction seen in Qwen3.8 27B (from $0.42 to $0.0249) and the largest output increase in DeepSeek V4 Flash Latest (from $0.32 to $1.25).
- The cheapest paid model by input cost is DeepSeek V4 Flash Latest at $0.012 per million input tokens.
- The largest context windows available are 2,000,000 tokens, offered by OpenRouter's Auto Router and Pareto Code Router, and x-ai's Grok 4.20 models.