Skip to content
Free AI model tracker

See which AI is best, at a glance.

Google's Gemini 4 Carbon is the smartest model this week. The best open model is MiMo-V2.6-Pro. Every number comes from human votes or independent tests, with a link to its source.

Arena data from Oct 28, 2026 · 490 models tracked

Top picks right now

One clear winner for each everyday question. Tap a card for the full story.

All categories →
Smartest
Gemini 4 Carbon
GoogleHigh
99SIscope score out of 100
Cod
Age
Rea
Mat
Cha
Ins
Top of the blind human-vote arena. Next best: Claude Opus 5.5.
 Best for coding
Claude Opus 5.5
AnthropicMax
100coding score out of 100
WebDev arena trend, recent updates
Wins the most head-to-head votes when people build real web apps.
 Best open model
MiMo-V2.6-Pro
XiaomiOPEN
88SIscope score out of 100
Human-vote score trend
Download the weights and run it yourself. The strongest open model we track.
 Best value
DeepSeek-V4.1-Flash-20260910
DeepSeekMaxOPEN
$0.16per 1M tokens · SIscope 85
This
Top 10
Top-10 average: $9 per 1M tokens
About 57× cheaper than the top 10, and still smarter than most.
Fastest
Speed tests coming
Coming soon
We are building our own hourly speed tests. No borrowed numbers, so nothing to show yet.

Smarts vs price

Up is smarter, left is cheaper. The dashed line links the best deals at each price.

ClosedOpen weightsBest deals
6065707580859095100$0.05$0.1$0.25$0.5$1$2$5$10$20List price per 1M tokens (3 in : 1 out, log scale) →Smarter →Claude Opus 5.5MiMo-V2.6-ProGLM-5.3-Flash

Smarts: SIscope score. Price: OpenRouter list price, blended 3 parts input to 1 part output. Only models with both are shown (80).

Open vs closed

Best closed model vs best open model in the blind human-vote arena.

49 ptsgap today · 28 pts on Aug 19, 2026
Aug 19Oct 28
Gemini 4 CarbonMiMo-V2.6-Pro

Arena text scores work like chess ratings: a 50-point lead means the leader wins roughly 57% of head-to-head votes.

What changed

Moves since the previous leaderboard update.

Gemini 4 Carbon takes #1New #1 in the text arena, ahead of Gemini 4 ArgonOct 28, 2026 · ArenaClaude Fable 5 climbs 1 placeNow #11 in text arenaOct 28, 2026 · ArenaMuse Spark 1.3 slips 3 placesNow #13 in text arenaOct 28, 2026 · Arena
Open models 49 points behindGap between best closed and best open was 35 at the previous updateOct 28, 2026 · Arena
Gemini 4 Carbon joins the coding board at #4WebDev arena (building web apps)Oct 28, 2026 · ArenaGPT-6.1 Terra gets a public price$1 in, $5 out per 1M tokens on OpenRouterOct 26, 2026 · OpenRouterKimi K3.1 Pro gets a public price$2 in, $10 out per 1M tokens on OpenRouterOct 23, 2026 · OpenRouter

Best for…

Pick what you want to do. We show the winner and the best open option.

See all →
CodingClaude Opus 5.5Best open: Kimi K3 WritingClaude Opus 5.5Best open: MiMo-V2.6-Pro MathGemini 4 CarbonBest open: MiMo-V2.6-Pro Images & visionClaude Fable 5Best open: GLM-5.3-Flash Agents & toolsClaude Opus 5.5Best open: MiMo-V2.6-Pro Open modelsMiMo-V2.6-ProTop open-weight pick Cheap & goodMiMo-V2.6-Pro$0.54 per 1M tokens Fast repliesComing soonOur own speed tests are on the way.

Top 10 overall

Ranked by SIscope score: human votes plus independent benchmark runs, 0 to 100. Each model at its best setting.

Full leaderboard →
1Gemini 4 CarbonHigh992Claude Opus 5.5High983Claude Fable 5.1Max964Gemini 4 ArgonHigh965Claude Sonnet 5.5xHigh946GPT-6 AstraMax937Qwen3.7-Max-Preview938GPT-6.1 SolMax939Claude Opus 5Max9210ERNIE 5.191

Bars start at 50 so the differences at the top are easy to see. Scores are percentiles: 90 means better than about 90% of tested models.