Mercury 2.5
#7 in LLM Inference & ServingInception · v2.5 · since 8. September 2026 (Launch-Ankündigung); Preview bereits 31. August 2026 laut Drittquelle · 2× · last seen Oct 21, 2026
32
Momentum
Mercury 2.5 is a diffusion large language model (dLLM) from Inception that generates text through parallel refinement instead of sequential decoding. According to Inception, it offers tunable reasoning, parallel tool calls, and schema-aligned JSON output. The context window is 260K tokens, and throughput is stated as 1,107 tokens/s on widely available NVIDIA GPUs.
Momentum trend
24.07.22.10.
Features
| Throughput/Latency | 1,107 tokens/s on widely available NVIDIA GPUs (vendor figure); 260K context |
| Platform | Inception API, Baseten, OpenRouter |
| Price | $0.20 / $0.75 per 1M tokens (input/output); launch promo $0.04 / $0.15 |
| Release Date | September 8, 2026 |
| Supported Models/Providers | Inception (own model) via Inception API, Baseten and OpenRouter |