Text Balanced
public-opt-in · serverless · 2026-09-05
840B
tokens
Rankingslive route intelligenceEmbeddings market
Embedding Model Rankings
Compare market share, route quality, task spend, benchmark fit, service tier, endpoint health, and GPU readiness in one practical ranking surface for customers, suppliers, and internal sales teams.
Market lens
Usage, spend, share
Routing signal
Quality, latency, cost
Buyer view
GPU-ready capacity
Weekly usage
38.4M
modeled requests ranked by route family
AI Token spend
$2.8M
credit share across models, apps, and routes
Ranking tracks
12
usage, spend, share, benchmarks, tools, images, apps, GPU
GPU readiness
18 lanes
serverless, reserved, dedicated, or private candidates
Window
7D live preview
Data scope
Embeddings route simulation
Export
model shortlist ready
Embedding Model Rankings
Vector search, memory, retrieval, classification, and app knowledge-base usage ranked by volume and cost.
weekly volume
166B
measurement
weekly embedded tokens
active routes
48+
Top Models
Weekly usage of embeddings routes across Aurona.
180B135B90B45B0
2025-06-30Aug 4Sep 8Oct 13Nov 17Dec 22Jan 26Mar 2Apr 6May 11Jun 15
subpage tracks
Each track changes the data table, cards, route config, and benchmark controls below.
Embeddings benchmarks
Benchmark scorecard
The scorecard converts raw model tests into route-ready ranking signals for product teams.
Quality89
human preference, instruction following, and route judge agreement
Retrieval fit84
RAG relevance, tool-call repair, and long-context recall
Latency83
p50 response time, queue depth, and stream start
Cost control84
AI Token margin, fallback waste, and cache efficiency
Reliability83
availability, retry success, and provider health
Embeddings performance
Latency, uptime, and capacity
p50 latency
92ms
availability
99.96%
retry save
87%
GPU reserve
L40S pool
stream start78%
provider health76%
fallback depth82%
cache hit rate91%
Embeddings economics
Task Spend
Credit consumption is grouped by workload so Aurona can rank not only models, but the businesses and APIs that create durable demand.
Aurona rankings console
Ranking views for models, routes, apps, and GPU lanes
Ranking sections are mapped into the Aurona business layer: AI Token spend, route quality, provider health, policy fit, latency, and private GPU readiness.
matched models
4
usage index
267
bench runs
142
Top Models
Animated 7D ranking for context length.
selected
DeepSeek
score
95
usage
96%
latency
640ms
Model Leaderboard
Click any row to update the config, benchmark panel, and route preview.
rankmodelscoreusagespendlatencyroute
By Task Spend
Share of AI Token credit spend by model family and workload type.
Benchmarks
Composite reasoning, instruction following, coding, vision, and safety scores.
Performance
Latency, availability, routing stability, and cost-efficiency signals.
Natural Languages
Multilingual assistants, translation, finance content, and regional applications.
Programming
Code generation, repo context, tool use, debugging, and PR review routes.
Context Length
Large-window models for documents, research, memory, and agent history.
Tool Calls
Function calling, structured JSON, multi-step tools, and production agents.
Images
Image input, vision reasoning, document images, and generated media pipelines.
Apps
Models ranked by downstream apps, app growth, and repeat customer usage.
GPU Ready
Models that can support reserved capacity, private deployments, or GPU-backed lanes.
Aurona Route Rankings
Rank models together with policy, cost, and GPU supply
Top Apps
App rankings connect model demand to the actual products that consume Aurona credits.
Meter-Safe Ranking Preview · simulated
Rank one native measure at a time.
Keep tokens, images, tool calls, media time, and GPU hours in separate evidence tracks. Aurona can normalize position inside a track without converting unlike units into a false universal score.
Native unit
tokens
Window
7D · through Sep 5 UTC
Proposed endpoint
POST /v1/rankings/native-meter-preview
meter_rank_00uotq20
Text Deliberate
public-opt-in · reserved-review · 2026-09-05
620B
tokens
Simulated evidence only. This preview does not publish market claims, convert native usage into billing, debit AI Tokens, rank private activity publicly, move traffic, or allocate GPUs.
Multimodal Evaluation Lab · simulated
Turn visual capability evidence into a route preview.
Hold candidates to the same prompt family, minimum case count, objective, and GPU capacity scope before comparing quality, generation time, and modeled AI Token use.
Proposed endpoint
POST /v1/evaluations/multimodal-route-preview
media_eval_00cyw2vn
Selected route preview
Image Precise
reserved-review · 12 shared cases
91%pass7.8sgeneration4.2AI Tokens
Image Precise
eligible · 12 cases · reserved-review
Image Fast
eligible · 12 cases · serverless
aurona/image-studio
excluded · capability mismatch
aurona/image-new
excluded · insufficient evaluation cases
Simulated evaluation evidence only. This preview does not run media jobs, publish benchmark claims, debit AI Tokens, move production traffic, or reserve GPUs.
Ranking Feed Lab · simulated
Make every ranking query explain its evidence.
Preview model, route, and app feeds with a declared measure, resolved window, visibility scope, stable version, portable dataset manifest, citation, AI Token context, and GPU capacity path.
Feed preview
/v1/datasets/rankings-preview?surface=all&scope=workspace-only&metric=ai-token-volume&window=7d
Version
aurona-ranking-preview-v3
Receipt
ranking_00av2q9o
Schema
aurona-ranking-feed/3
Format
JSON manifest
Data through
Sep 1 · UTC
Resolved window
7 days
Measure
modeled AI Token volume index
Balanced Text
models · public-opt-in · 124,000 modeled requests
84
serverless
Agent Route
routes · public-opt-in · 108,000 modeled requests
71
reserved-review
Private Agent Route
routes · workspace-only · 91,000 modeled requests
63
dedicated-review
Portable citation
Aurona.ai, "Aurona Ranking Preview" (aurona-ranking-preview-v3), modeled AI Token volume index, trailing 7 days through 2026-09-01 UTC, workspace-only scope, simulated.
Origin: simulated Aurona route evidence. This manifest identifies the dataset and methodology; it does not state reuse rights.
Simulated ranking evidence. Growth compares adjacent complete windows and signals adoption, not model quality. This preview does not publish traffic, change visibility, debit AI Tokens, or reserve GPU capacity.