Mercury 2.5 ranks #170 of 228 on the public leaderboard with a score of 33.38/100. It does not yet have enough sourced coverage for a verified position.
Mercury 2.5 is a proprietary model with a 260K 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-2.5`, with 260K context, tunable reasoning, parallel tool calls, and schema-aligned JSON. The launch also lists Baseten and OpenRouter access.
Inception released Mercury 2.5 on September 8, 2026. The launch chart supplies provider-reported GPQA Diamond, IFBench, long-context reasoning, SciCode, DeepSearchQA at ten tool calls, Omniscience accuracy and non-hallucination, and Tau3Bench Telecom results. TerminalBench lacks a version; the GDPval bar uses an unexplained percentage scale. Neither is mapped. Benchmark values are separate from Mercury 2 and Mercury 2.5 Preview.
Mercury 2.5 sits in the Mercury 2.5 family with Mercury 2.5 Preview. Its explicit predecessor is Mercury 2. 6 of 764 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.
Its strongest eligible category is Instruction Following at #60, while its lowest eligible position is Knowledge at #139. a well-rounded choice across a range of tasks.
Model research
The launch rates are temporary
The models page lists standard rates of $0.20 per million input tokens, $0.02 per million cached input tokens, and $0.75 per million output tokens. The 80% launch discount reduces them to $0.04, $0.004, and $0.15. Inception does not state an end date on that page.
The first-party quick start uses the API model ID mercury-2.5. Mercury 2.5 supports adjustable reasoning, parallel tool calls, and JSON constrained to a schema. Mercury 2 remains supported for existing customers.
What the launch chart establishes
Inception reports 79% on GPQA Diamond, 77% on IFBench, and 38% on SciCode. The chart also reports 34% on DeepSearchQA with a ten-tool-call budget. These values belong to Mercury 2.5 and are not copied to Mercury 2 or the earlier Preview record.
The chart’s 96% Tau3Bench result is labeled Telecom in its source note. It should not be compared directly with the older Mercury 2 Banking result. Its TerminalBench bar does not identify a benchmark version, so we leave that result outside the versioned ledger until the setup is documented.
Measure the complete request
Inception reports 1,107 output tokens per second on NVIDIA GPUs. That is a provider throughput claim, separate from the hosted runtime measurements. It does not establish time to first token or the full response latency of a voice application.
Mercury Voice and Mercury Router were announced as separate previews. Their latency and routing claims do not become Mercury 2.5 specifications.