Mercury 2.5 reaches 1,107 tokens per second

Inception released Mercury 2.5, reporting 1,107 output tokens per second on widely available NVIDIA GPUs, a 260K-token context window, and standard API pricing of $0.20 per million input tokens and $0.75 per million output tokens. Inception’s release says the diffusion model adds tunable reasoning, parallel tool calls, and schema-aligned JSON; its speed and quality comparisons remain vendor-reported. Latency-sensitive builders can put Mercury 2.5 in a bounded routing bake-off, using OpenAI’s 750-token-per-second Ultrafast preview as a reminder to measure quality, rate limits, and post-discount price before switching.