How they use Nebius and Tavily
Artificial Analysis assessed Nebius AI Studio endpoints for quality, speed and price. Nebius’s November 2024 announcement reported assessments across all AI Studio endpoints and quoted co-founder George Cameron on their competitive pricing. Cameron specifically cited Llama 3.1 405B prices of $1 per million input tokens and $3 per million output tokens.
Its provider benchmarks also include Nebius endpoints for Nemotron 3.5 Super VL and Nemotron 3 Super 120B A12B (Reasoning). It also benchmarks Beam 1.0 in NVFP4, Muse Spark Open (Xhigh) in FP4 and MiniMax-M3.1-Flash through Nebius. Its Nebius provider page tracks 38 models, including DeepSeek V4.1 Flash 1017 (max), GLM-5.3 (max) in FP4 and GLM-5.3-Flash.
Its GLM-5.3 (max) comparison reports 239 output tokens per second and 1.49-second first-chunk latency for Nebius. It also benchmarks Nebius’s DeepSeek V4 Pro 0813 (max) endpoint alongside eight other API providers. Cache-hit measurements use multi-turn agentic coding sessions, with representative rates reported from recent measurements. Artificial Analysis continuously benchmarks Nebius and displays historical performance trends.
For the deprecated Nemotron 3 Super model, performance benchmarking continues only for the default 10,000-input-token workload. Other workload results are historical and no longer updated.
Artificial Analysis also evaluated Tavily Search with basic search depth for its Search Index. Its Stirrup agent harness paired a fixed answer model with web_search and web_fetch tools, allowing up to 25 turns per task. The answer model was GPT-5.6 Luna (medium). Search responses used snippet content with up to 10 results, and known contamination sources were filtered.
Search-provider errors were retried until success, while recoverable page-fetch errors remained in published results. The index combines DeepSearchQA, BrowseComp and AA-Omniscience results. Tavily Search (basic) scored 66 at the August 2026 launch. Parallel’s comparison reports Tavily’s component scores as 74 on DeepSearchQA F1, 59 on BrowseComp and 64 on AA-Omniscience, with 36.3 seconds per task.
Sources
14 sourcesDeepSeek V4 Pro 0813 (max): API Provider Performance Benchmarking & Price Analysis | Artificial Analysis
Listed 3 Nov 2026 · artificialanalysis.ai · Web page
Artificial Analysis benchmarks and tracks Nebius’s DeepSeek V4 Pro 0813 (max) endpoint among nine API providers. Its comparison reports 170 output tokens per second and 1.61-second first-chunk latency for Nebius.
Open original sourceAnnouncing the Artificial Analysis Search Index: Same Agent, Different Search
Published 18 Aug 2026 · Artificial Analysis · Web page
Artificial Analysis evaluated Tavily Search (basic) using GPT-5.6 Luna (medium) in Stirrup. Search responses used snippet content with up to 10 results and known contamination sources filtered out.
Open original sourceNebius - Intelligence, Performance & Price Analysis | Artificial Analysis
Listed 28 Oct 2026 · artificialanalysis.ai · Web page
Artificial Analysis tracks 38 models offered through Nebius, including DeepSeek V4.1 Flash 1017 (max), GLM-5.3 (max) in FP4 and GLM-5.3-Flash.
Open original sourceNemotron 3 Super: API Provider Performance Benchmarking & Price Analysis | Artificial Analysis
Listed 27 Oct 2026 · artificialanalysis.ai · Web page
Artificial Analysis benchmarks Nebius among three API providers for Nemotron 3 Super 120B A12B (Reasoning). The model is deprecated, with performance benchmarking continuing only for the default 10,000-input-token workload. Results for other workloads are historical and no longer updated.
Open original sourceCompany context
artificialanalysis.aiArtificial Analysis, operated by Artificial Analysis, Inc., provides independent benchmarks and analysis to help developers and businesses choose AI technologies. Its website compares language models, coding agents, image, video and speech models, inference services and AI chips across capability, speed and cost. Its products include public leaderboards, the Artificial Analysis Intelligence Index and model comparison tools. The company also sells enterprise benchmarking and insights subscriptions and conducts private custom benchmarking for companies building AI models and infrastructure.
Artificial AnalysisRecent updates
24 updatesWe are launching the Artificial Analysis Endpoint Accuracy Index v1.1, a minor version upgrade alongside new coverage for GLM-5.3 Flash and Muse Spark Open. Across the 103 serverless endpoints we benchmarked, 24 score statistically below our self-hosted reference.
We are launching the Artificial Analysis Endpoint Accuracy Index v1.1, a minor version upgrade alongside new coverage for GLM-5.3 Flash and Muse Spark Open.
artificialanalysis.ai announcementIntroducing the Artificial Analysis Decision Index v2
Introducing the Artificial Analysis Decision Index v2: a new proprietary evaluation, updated benchmarks, and a headline metric that rewards model calibration
artificialanalysis.ai announcementArtificial Analysis published its VIALS-AA benchmark launch article
Artificial Analysis introduced VIALS-AA, its implementation of Handshake AI’s benchmark for testing whether language models can interpret scientific images. The article reports results as of November 6, 2026.
Artificial Analysis