Semifly
Semifly · Tokens & LLM Models

Large Language Models, made easy to compare, download, and run

Track the latest models, compare frontier and open-source LLMs, and download open-weight models — then run any of them on Semifly with tokens, GPU servers, and AI Foundry.

Genealogy

The whole family tree

Every major lab grouped by camp — US vs China, proprietary vs open — with each family’s key versions and years. The China open-weight branch is by far the busiest.

US proprietaryUS/EU openChina openChina proprietary

Key versions per lab, hand-curated as of 2026-06. Refreshed as new flagships ship.

Quality × Price

The whole field on one chart

Every model placed by quality (Artificial Analysis Intelligence Index) and input price — top-left is the value sweet spot. As of 2026-06-18.

Proprietary (API)Open-weight (self-host)Bubble size = context window
Capability over time

How the field got here — and how open caught up

LMArena Elo of the #1 proprietary model (blue) vs the #1 open-weight model (orange), 2023–2026. Open-weight nearly drew level in early 2025; since then the open #1 has been almost entirely Chinese — DeepSeek, Qwen, GLM, Kimi.

Proprietary #1Open-weight #1 (diamond = China)

Source: LMArena (Arena) Elo via BenchLM leaderboard history. Blue tracks the frontier milestones; orange is the open-weight #1 over time.

Adoption

Who is actually being used — and who gets paid

OpenRouter routes about 25T tokens a week across 8M+ developers. By real token volume, Chinese open-weight models dominate the usage charts — yet premium US models still capture most of the dollars.

China vendorOverseas

Top models · weekly tokens

By vendor · top-10 aggregate

The dollar–token split: China-origin models take 45%+ of tokens, while Anthropic holds ~12% of tokens but ~46% of dollar spend through premium pricing. Source: OpenRouter rankings (live-scraped) + market reporting. Per-language and per-use-case breakdowns are chart-only on OpenRouter and not yet scrapable.

Latest in LLMs

What’s new in large models

Updated July 2026 · refreshed regularly

Jul 24, 2026Model rankings

Anthropic claims its new Claude Opus 5 delivers near-Fable 5 performance at half the token price

Anthropic's new flagship model Claude Opus 5 posts top scores in coding and knowledge work at half of Fable 5's token rates. On ARC-AGI-3, a benchmark for novel…

The Decoder →
Jul 24, 2026Open source

Microsoft's open-weight AI push is so obviously an Azure play it hurts

Microsoft, along with Meta, Nvidia, and more than 20 other companies, is pushing for open-weight AI models in an open letter. The strategic logic is simple: the more…

The Decoder →
Jul 24, 2026Model rankings

Sakana claims its AI model router Fugu Ultra v1.1 now beats Fable 5 without even including it in the pool

Sakana AI has updated its Fugu Ultra AI router to version 1.1, claiming gains of up to 7.9 points over v1.0. Independent verification doesn't exist yet. The update adds…

The Decoder →
Jul 24, 2026Release

Introducing Claude Opus 5

Introducing Claude Opus 5 I've been offline kayaking with sea otters for much of today so I haven't had a chance to put Anthropic's new model Claude Opus 5 through its…

Simon Willison →
Jul 19, 2026Release

AI Mania Is Eviscerating Global Decision-Making

AI Mania Is Eviscerating Global Decision-Making Here's an entertaining perspective from Nik Suresh on the AI mania that is overwhelming the large companies that he…

Simon Willison →
Jul 16, 2026Release

Firefox in WebAssembly

Firefox in WebAssembly This is absurdly cool: Puter compiled Firefox to WebAssembly such that the whole browser runs in another browser. Here's my blog, running in…

Simon Willison →
Jul 15, 2026Release

OpenAI is now using AI to attack its own AI, and it's working better than humans ever did

OpenAI's internal GPT-Red model finds successful attacks in 84 percent of test scenarios through self-play training. Human red teamers manage just 13 percent. The…

The Decoder →
Jul 15, 2026Release

GPT-5.6 Sol reportedly disproves a 30-year-old statistics conjecture in 90 minutes after humans couldn't crack it

A University of Pennsylvania statistics professor used OpenAI's GPT-5.6 Sol Pro to disprove a central open conjecture about the Benjamini-Hochberg method in roughly 90…

The Decoder →
View all news & China watch →

Run any of these on Semifly

Tokens & API

Access hosted models through a simple, metered token API.

Get API access →

GPU servers

Buy or lease Supermicro GPU systems to self-host open-weight models.

Browse GPU servers →

AI Foundry

Managed compute for training, fine-tuning, and inference.

Explore AI Foundry →