Skip to content
overshoot

Every vision model, measured for real time

Overshoot runs every Vision Language Model on live video and benchmarks what actually ships: capability, speed, latency under the 200 ms bar, and cost, across every provider and region.

Explore leaderboardsHow we measure
models measured
13models measured
providers
11providers
regions
8regions
Real-time bar
200 msreal-time bar

Latest

All updates

Intelligence Index v2.1

Structured Extraction v2 and a refreshed Video QA suite are in. We re-ran every model on the Overshoot API on live video.

Jun 2026

Real-time readiness

A new latency track that flags which models clear the 200 ms end-to-end bar and can close the loop on live video.

Jun 2026

Who leads right now

Intelligence

VLM Intelligence Index

86Claude Opus 4.6 · Anthropic
Intelligence leaderboard

Speed

Output tokens per second

429 tok/sGemma 4 26B A4B · Google
Speed leaderboard

Real-time

Models under the 200 ms end-to-end bar

6of 13 models
Latency & real-time

Cost / task

USD per Index task · Lower is better

$0.03Qwen3.6 27B · Alibaba (Qwen)
Price & cost

Top models

Full leaderboard
#ModelDeveloperIndexSpeed$ / 1MReadiness
1Claude Opus 4.6Anthropic8655 tok/s$6.93Batch
2GPT-5.4OpenAI8591 tok/s$6.68Batch
3Claude Sonnet 4.6Anthropic8395 tok/s$6.32Batch
4Gemini 3.1 ProGoogle83106 tok/s$5.83Batch
5Qwen3.6 35B A3BAlibaba (Qwen)81406 tok/s$0.12Batch

Intelligence isn't free

Cost per task is the weighted average USD to run one Intelligence Index task at each model's reference provider. Reasoning models are marked with a bulb.

What we measure

Capability

What each model reads off a frame: OCR, documents and charts, scene reasoning, video QA, grounding.

Speed

Sustained output tokens per second on live video.

Latency

Time to first token and end-to-end response under load, across eight regions.

Cost

Blended price per 1M tokens, cost per task, and per-frame math.

Real-time

Which models clear the 200 ms bar on live camera and screen.

Pick the model that sees and acts in real time

Explore leaderboardsCompare models