Language model rankings
Quality, speed, price and usage — kept as four separate axes. Usage is popularity, not quality: a cheap fast model wired into one busy application outranks a more capable one used sparingly.
Last synced 2026-10-13 · newest measurement 2026-10-12 · 3 sources(1 stale)
Quality
as of 2026-10-12Marsel score — our composite of the quality metrics every active source publishes. Higher is better.
Showing 10 models; 277 excluded for fewer than two measured inputs — see the formula.
At least one source has not refreshed within twice its stated cadence. The values above may not be current.
Top 10 of 58 measured — the rest are on the tab below.
View data
| Model | Provider | Marsel score |
|---|---|---|
| Gemini 3.8 Flash (high) | Google DeepMind | 96.6 |
| Gemini 3.7 Flash (high) | Google DeepMind | 95.2 |
| Gemini 3 1 Pro Preview | 95.2 | |
| Claude Opus 5 (max) | Anthropic | 95.1 |
| Claude Opus 4.7 (max) | Anthropic | 94.7 |
| Gemini 3.6 Flash | Google DeepMind | 94.4 |
| Glm 5 3 Max | zai | 92.5 |
| Claude Fable 5 (high) | Anthropic | 91.9 |
| Gemini 3.5 Flash | 91.8 | |
| Deepseek V4 1 Flash Max | DeepSeek | 91.6 |
Speed
Tokens generated per second once output has started. Varies by provider and load. Higher is better.
No active source publishes output speed yet. The axis is here because it is one a reader comes to compare — the numbers arrive when a source that measures it is switched on.
Cost
as of 2026-10-12Published price per 1M output tokens. LOWER is better here — the leftmost column is the cheapest, and therefore the shortest.
Showing 10 models; 191 excluded for no published price from an active source.
Top 10 of 144 measured — the rest are on the tab below.
View data
| Model | Provider | Output $/1M |
|---|---|---|
| Apodex: Apodex 1.1 Mini (free) | Apodex | $0.00 |
| Cohere: North Mini Code (free) | Cohere | $0.00 |
| Dots Studio: Dots3-Note Preview (free) | Dots Studio | $0.00 |
| Free Models Router | Openrouter | $0.00 |
| Google: Lyria 3 Clip Preview | $0.00 | |
| Google: Lyria 3 Pro Preview | $0.00 | |
| Google: Gemma 3 4B | $0.10 | |
| Google: Gemma 3n 4B | $0.12 | |
| Amazon: Nova Micro 1.0 | Amazon | $0.14 |
| Cohere: Command R7B (12-2024) | Cohere | $0.15 |
Most used
as of 2026-10-12Tokens actually routed through these models. This is POPULARITY, not quality — a cheap model wired into one busy product outranks a better one used sparingly.
Showing 10 models; 306 excluded for no usage volume from an active source.
Top 10 of 29 measured — the rest are on the tab below.
View data
| Model | Provider | Usage (tokens) |
|---|---|---|
| Deepseek V4 1 Flash 20260910 | DeepSeek | 6.0T |
| Glm 5 3 Flash 20260826 | Z.ai | 2.0T |
| Claude Opus 5 5 20260921 | Anthropic | 788.5B |
| Glm 5 3 20260816 | Z.ai | 652.1B |
| Claude Sonnet 5 5 20260928 | Anthropic | 374.3B |
| Deepseek V4 Flash 20260731:Free | DeepSeek | 340.8B |
| Gemini 3 8 Flash 20260902 | 297.0B | |
| Claude Haiku 5 5 20261007 | Anthropic | 253.0B |
| Gemini 3 Flash Preview 20251217 | 190.7B | |
| Claude Opus 5 20260723 | Anthropic | 131.1B |
Sources · Methodology · Marsel measures nothing itself — every figure is a published measurement, ingested with its source and date.