vakati-tech.ai

// llm_stats · November 2026

LLM statistics

Live numbers on the language-model market — how many models there are, what they cost, and what quality costs — computed from published prices and benchmark runs every 15 minutes. Free to cite: each figure has its own link and a ready-made citation.

As of November 20, 2026 JSON CSV Embeddable widgets

Market

LLMs available through one API

354

Distinct models listed on OpenRouter from 60 providers, counting free and batch tiers as the same model.

New models in the last 30 days

42

Listings added in the past 30 days; 137 in the past 90.

Price

Median price per 1M tokens

$0.850

Median blended list price across 345 paid models, weighting input 3:1 over output.

Median price of the 10 best-scoring models

$7.00

Blended list price, median of the top 10 on LiveBench overall — 8.2× the median of all paid models.

Free model listings

14

Models listed at $0 for input and output, including free tiers of paid models. Free tiers carry tighter rate limits.

Quality & value

Highest LiveBench overall score

87.4

Gemini 4 Neon, at $6.00 per 1M blended tokens.

Cheapest model scoring 80+ overall

$4.00

GPT-6.2 Sol, scoring 80.3 on LiveBench overall.

Price spread at top-tier quality

5×

Among the 7 models within 5 points of the leader, list prices run from $4.00 to $20.00 per 1M tokens.

Lowest measured cost per point

$0.0013

GPT-6 Luna, scoring 60.4 — from what LiveBench actually spent per task, not list price.

API cost of a heavy coding-agent month

$668–$6,540

Across the 5 best coding models: GPT-6.1 Terra to Claude Fable 5.1, for 20K requests of 40K tokens in and 4K out with 70% cached input, at list price.

Capabilities

Largest context window

2M

SpaceXSI: Grok 4.20 Multi-Agent. 140 models (40%) accept 1M tokens or more.

Models with tool calling

84%

296 of 354 models, as declared by the provider.

Models with extended reasoning

70%

247 of 354 models, as declared by the provider.

Models that accept images

65%

229 of 354 models, as declared by the provider.

Method and reuse

  • Sources — list prices, context and capabilities from OpenRouter; scores and measured spend from LiveBench release 2026-11-06.
  • Counting — free and batch listings of a model count as that model; “blended” price weights input 3:1 over output.
  • Reuse — free to cite and republish with attribution (CC BY 4.0). Numbers move as prices and releases change, so cite the date.