07 / notes
How to read this profile
The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.
We track Mercury 3, but the public leaderboard excludes this profile until enough non-generated benchmark coverage is available. Only published rows appear above.
Mercury 3 is a proprietary model with a 1M context window. It uses an explicit reasoning mode, which can improve complex problem solving while adding latency and token use.
Available through the Inception API as mercury-3 with a 1M-token context. Inception also announces OpenRouter, Baseten, and Vercel access; the OpenRouter public model list had not listed the exact route when reviewed on October 19, 2026. First-party launch pricing is 50% off: $0.10/M input, $0.01/M cached input, and $0.35/M output, with no stated promotion end date.
Its explicit predecessor is Mercury 2.5. 8 of 686 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.
Model research
The chart uses medium reasoning
Inception reports 89.23 on GPQA Diamond, 76.84 on IFBench, 60.35 on MultiChallenge, 75.50 on MRCR v2, and 46.31 on DeepSWE v1. Its agent results are 30.24 on tau3-bench Banking, 55.18 on BrowseComp+, and 71.33 on DeepSearchQA.
The chart does not disclose exact harness revisions, trial counts, or tool budgets. MRCR context lengths and needle count are also unspecified. We keep Banking separate from Telecom reports and BrowseComp+ separate from original BrowseComp. The SWE-Bench Atlas task split is unnamed, so its 51.08 is not mapped to a specific Atlas benchmark.
Launch pricing is half the standard rate
The first-party models page lists launch rates of $0.10/M input, $0.01/M cached input, and $0.35/M output. Standard rates are $0.20, $0.02, and $0.70 respectively. Inception states no promotion end date.
The models page demonstrates mercury-3 in an OpenAI-compatible chat-completions request. The launch also announces OpenRouter, Baseten, and Vercel access. The public OpenRouter model list had not listed the exact route when we checked on October 19, 2026.
The speed figure is a provider claim
Inception reports 1,068 output tokens per second on NVIDIA GPUs. That figure does not establish time to first token or full application latency. We keep it separate from independent runtime measurements and do not inherit Mercury 2.5 measurements.