Skip to content
Kumo

Models

AI model benchmarks and rankings

Artificial Analysis’ independent measurements and real OpenRouter popularity in one place: who is smarter, faster and cheaper per task, which coding agents and image generators lead — and which of them Kumo sells.

Who leads right now

The best models on each key metric: intelligence, speed, latency, cost per task, context window, coding agents and image generation.

analysis

Which models play the best chess — and why it matters

A fast DeepSeek model just beat Claude, GPT and Kimi over the board. Chess Elo looks like a proxy for hard algorithmic reasoning — here’s the debate, and the practical takeaway for picking a model.

analysis

Why an LLM is a risky fact-checker — and how not to get fooled

A model has no opinion of its own: it interpolates over its training data and defends its early mistakes. Here’s why frequency beats truth, why a chat gets stubborn mid-conversation, and a practical protocol for validating facts with AI.

pricing

Why Kumo’s rate lands 30–50% below list — and how to check it

The lower rate is real, not a teaser. Here’s what’s guaranteed, what moves the number, and how to verify it yourself.

Intelligence

Artificial Analysis Intelligence Index

higher is better
  1. 158
  2. 256
  3. 356
  4. 456
  5. 554

The smartest models. Claude Opus 5.5: 58. Then Claude Opus 5.5 (Xhigh) and Claude Sonnet 5.5.

Output speed

Output tokens per second

higher is better
  1. 11966
  2. 2987
  3. 3925
  4. 4410
  5. 5385

The fastest models. Celeris-1: 1966. Then K2 Horizon 375B A23B and Mercury 2.5.

Cost per task

Average cost of one index task, USD

lower is better
  1. 1$0.005
  2. 2$0.006
  3. 3$0.008
  4. 4$0.008
  5. 5$0.01

Cheapest per task. GPT-6 Luna (low): $0.005. Then Granite 4.2 3B and Ministral 3 3B.

anthropic

Claude Opus 5.5: what changed and how to use it on Kumo

Anthropic shipped Claude Opus 5.5 — Fable 5.1's level for $4 and $20 per million tokens. Kumo pricing, the benchmarks, and two checks that prove the original model is answering.

Coding Agent Index

Artificial Analysis Coding Agent Index

higher is better
  1. 1Sonnet 5.5 (max)68
  2. 2Claude Opus 5.5 Max + GPT-6 Luna High68
  3. 3Muse Wonder Full (max)68
  4. 4Opus 5.5 (max)66
  5. 5Gemini 4 Argon64

The best coding agents. Claude Code (Sonnet 5.5 (max)): 68. Then Devin Fusion CLI (Claude Opus 5.5 Max + GPT-6 Luna High) and Muse Code (Muse Wonder Full (max)).

Time per task

Average agent wall time per task

lower is better
  1. 1Sonnet 5.5 (low)6.3 min
  2. 2Sonnet 5.5 (medium)8.5 min
  3. 3GPT-6.1 Sol (low)8.6 min
  4. 4GPT-6.1 Sol (medium)10.9 min
  5. 5Haiku 5.5 (low)11.6 min

Fastest through a task. Claude Code (Sonnet 5.5 (low)): 6.3 min. Then Claude Code (Sonnet 5.5 (medium)) and Codex (GPT-6.1 Sol (low)).

Cost per task

Average API cost per task, USD

lower is better
  1. 1DeepSeek V4 Flash 0731 (max)$0.09
  2. 2Haiku 5.5 (low)$0.1
  3. 3GPT-6 Luna (max)$0.2
  4. 4Haiku 5.5 (medium)$0.2
  5. 5DeepSeek V4 Pro 0813 (max)$0.2

Cheapest per task. Codex (DeepSeek V4 Flash 0731 (max)): $0.09. Then Claude Code (Haiku 5.5 (low)) and Codex (GPT-6 Luna (max)).

comparison

Best AI Vibe-Coding Tools in 2026: An Honest Comparison

We compare the best AI coding tools of 2026 — Claude Code, Cursor, Cline, OpenCode and more: who each one is for, what it costs, and how to choose.

Text to image

Blind pairwise comparisons on the Artificial Analysis arena

higher is better
  1. 11220
  2. 21207
  3. 31186
  4. 41174
  5. 51167

Best at generating. GPT Image 2.5 Sunburst (max): 1220. Then GPT Image 2.5 Flare (max) and GPT Image 2 (high).

Image editing

Blind pairwise comparisons on the Artificial Analysis arena

higher is better
  1. 11199
  2. 21182
  3. 31152
  4. 41152
  5. 51144

Best at editing. GPT Image 2.5 Sunburst (max): 1199. Then GPT Image 2.5 Flare (max) and MAI-Image-2.6.

comparison

Kumo and OpenRouter: the honest comparison

Both are OpenAI-compatible gateways to many models. Here’s how their operating models differ, including cases where a broader public catalog matters more.

pricing

Why Kumo’s rate lands 30–50% below list — and how to check it

The lower rate is real, not a teaser. Here’s what’s guaranteed, what moves the number, and how to verify it yourself.

Popularity

Share of OpenRouter tokens over the last week

higher is better
  1. 122%
  2. 211%
  3. 38%
  4. 47%
  5. 55%

Most used this week. DeepSeek V4.1 Flash 1017: 22%. Then Step 5 Preview and GLM-5.3-Flash.

Weekly climb

Gain in share against the week before, percentage points

higher is better
  1. 1+8 pp
  2. 2+1 pp
  3. 3+1 pp
  4. 4+1 pp
  5. 5+1 pp

Climbing fastest. Step 5 Preview: +8 pp. Then GLM-5.3 and Claude Opus 5.5.

comparison

Kumo and OpenRouter: the honest comparison

Both are OpenAI-compatible gateways to many models. Here’s how their operating models differ, including cases where a broader public catalog matters more.

claude

Low-priced Opus and Fable tokens for developers who code intensively

Opus 5 and Fable 5.1 prices on Kumo, Anthropic API comparisons and high-volume calculations. No weekly subscription quotas; card or crypto payments.

Intelligence

Artificial Analysis Intelligence Index

higher is better
  1. 158
  2. 256
  3. 356
  4. 456
  5. 554

Output speed

Output tokens per second

higher is better
  1. 11966
  2. 2987
  3. 3925
  4. 4410
  5. 5385

Latency

Time to the first token of the answer

lower is better
  1. 10.2 s
  2. 20.2 s
  3. 30.2 s
  4. 40.3 s
  5. 50.3 s

Cost per task

Average cost of one index task, USD

lower is better
  1. 1$0.005
  2. 2$0.006
  3. 3$0.008
  4. 4$0.008
  5. 5$0.01

Context window

How many tokens the model takes in one request

higher is better
  1. 110M
  2. 22M
  3. 32M
  4. 42M
  5. 52M

Coding agents

Artificial Analysis Coding Agent Index

higher is better
  1. 1Sonnet 5.5 (max)68
  2. 2Claude Opus 5.5 Max + GPT-6 Luna High68
  3. 3Muse Wonder Full (max)68
  4. 4Opus 5.5 (max)66
  5. 5Gemini 4 Argon64

Images

Blind pairwise comparisons on the Artificial Analysis arena

higher is better
  1. 11220
  2. 21207
  3. 31186
  4. 41174
  5. 51167

data as of October 17, 2026Data sources