Compare AI Models:Benchmarks, Rankingsand Prices
Live benchmarks across 334 AI models, plus MCP servers, comparisons, news and beginner-to-deep guides — everything AI, on evidence rather than hype.
The Companies Building AI
We track 334 models across coding, image, video and audio generation. Here are ten of the companies behind them.
The only provider on the roster with models in all four categories. Flagship: Gemini Omni Flash.
OpenAI
Covers all four categories, from the GPT-5 series to Sora and its audio models.
Alibaba
The Qwen family spans coding, image and video, with open weights across much of the range.
SpaceXSI
Grok across coding and image, plus video generation. Formerly xAI; x.ai now publishes as SpaceXSI.
Anthropic
Text and coding only, and the top of the LMArena text board is almost entirely Claude.
Meta
Muse Spark and the Llama line, across coding, image and video.
MiniMax
One of the few providers pairing text models with both video and audio generation.
Black Forest Labs
The FLUX family, image generation extended into video.
ByteDance
Seedream and Seedance, covering image and video generation.
ElevenLabs
Audio generation specialist - speech synthesis and voice models.
Ranked by LMArena Elo
The top 25 of 277 models with a published LMArena Elo, highest first — how they place against each other in blind head-to-head comparisons. Each bar carries its 95% range; 2 models' ranges reach the leader's, so the order between them is not evidence of a difference. The full board is on the AI model rankings page.
- Gemini 4 CarbonGoogle DeepMind · high effort1528±10
- Gemini 4 ArgonGoogle · high effort1520±7
- Claude Opus 5.5Anthropic · high effort1509±7
- Muse WonderMeta · max effort1507±7
- Claude Fable 5.1Anthropic · max effort1504±6
- Claude Fable 5Anthropic · high effort1504±4
- Claude Opus 4.6Anthropic · high effort1503±3
- Claude Opus 4.7Anthropic · high effort1501±4
- Claude Opus 4.6Anthropic · default effort1496±3
- Gemini 3.8 FlashGoogle · high effort1496±4
- Muse Spark 1.3Meta · max effort1495±5
- Claude Opus 4.7Anthropic · default effort1494±4
- Muse Spark 1.2Meta · xhigh effort1492±7
- Claude Opus 5Anthropic · high effort1492±4
- Claude Opus 5Anthropic · max effort1491±5
- Muse Spark 1.1Meta1490±4
- Gemini 3.7 FlashGoogle · high effort1490±5
- Muse SparkMeta1489±6
- Kimi K3Moonshot AI · max effort1488±5
- Gemini 3.1 ProGoogle1487±3
- Gemini 3 Pro (Retired)Google1485±4
- GPT 5.6 SolOpenAI · xhigh effort1485±4
- Gemini 3.6 FlashGoogle · high effort1483±5
- Qwen3.8Alibaba · max effort1482±5
- Claude Opus 4.8Anthropic · high effort1482±4
Head to head
Two models.
Four measurements.
Every bar below is the gap between them, drawn from the centre towards whichever model leads. Same figures the comparison tool uses — straight off the roster.
- 1528LMArena Elo1520Gemini 4 Carbon is 0.5% ahead on LMArena Elo.
- $2Input per 1M$2Both $2 per million input tokens.
- $10Output per 1M$10Both $10 per million output tokens.
- 1MContext window1.05MGemini 4 Argon takes 4.9% more context in one request.
Neither wins outright. Gemini 4 Carbon takes 1, Gemini 4 Argon takes 1 — which of those rows matters is your call.
Open this comparisonAI News & Trending
Coverage of the latest breakthroughs, model launches, and developments in artificial intelligence.
Learning path
Want to learn AI?
Start here.
Four guides, in order, each one assuming the last. No jargon and nothing taken for granted.
36 minutes end to end
Browse by Category
Guides and articles, grouped by topic.
AI News In Your Inbox
Once a week, the week’s AI news from AIblogly: up to ten stories, each with a short summary and a link. No email in a week with no news.
- Weekly
- Up to ten stories
- Unsubscribe in one click
No spam. Unsubscribe anytime.