OpenAI Cuts GPT-5.6 Sol Pricing in Half to $2.50/$15 per Million Tokens

OpenAI’s flagship GPT-5.6 Sol model has dropped 50% in price, now listed at $2.50 per million input tokens and $15 per million output tokens on OpenRouter. Released July 9, 2026 with a February 2026 knowledge cutoff and a 1M-token context window, the model is pitched at complex reasoning and agentic coding — specifically long-horizon, multi-step, command-line work. The cut lands the frontier model in more competitive territory for developers running high-volume coding and agent workloads.

Because the model is served through OpenRouter, it runs across multiple hosts — Azure US/EU, OpenAI direct, and Amazon Bedrock — with requests routed by mode (Balanced, Nitro for speed, or Exacto for tool-calling accuracy). That aggregation is a reliability play: OpenRouter reports 99.65% availability with cross-provider failover versus 97.55% without it, plus roughly 46 tokens/sec throughput and 3.28s median latency at the best provider. Benchmark scores cluster around 90–92% on GPQA Diamond and 70–80% on TAU-Bench depending on the host.

The more telling signal is what’s actually consuming the model. The top production traffic comes from agentic tools — Nous Research’s open-source, memory-persistent Hermes Agent (201B tokens), Anthropic’s Claude Code (61.5B), and the open-source OpenClaw agent (52.7B). Cheaper token pricing on a strong long-horizon model directly lowers the cost of running these autonomous, tool-using agents, which is where a growing share of real workloads now sits.