Curated, every major LLM as of mid July 2026 and how they currently rank

[Odd Arrow]

A snapshot of where the frontier actually stands right now, since this changes fast enough that lists go stale within weeks. Based on Artificial Analysis Intelligence Index scores, SWE-bench and GPQA Diamond results, and current market share data as of mid July 2026

Overall frontier leader, Claude Opus 4.8 currently tops the Artificial Analysis Intelligence Index at 61.4, edging out GPT-5.6, with particularly strong scores in coding, SWE-bench Pro 69.2 percent, and computer use agent tasks, Online-Mind2Web 84 percent

Best for coding specifically, Claude Fable 5 leads frontier coding benchmarks at roughly 95 percent on SWE-bench Verified by some measures, having just returned to public availability on July 1 after an 18 day suspension tied to a US export control order, notably the first frontier model ever switched off and back on for regulatory reasons rather than a technical or safety issue

Default everyday chat model, GPT-5.6, specifically the Sol variant, became ChatGPT's new default on July 9, replacing GPT-5.5, and is tuned for harder math, science and cybersecurity reasoning, alongside cheaper Terra and Luna variants aimed at everyday use and budget conscious deployments respectively

Best for hardest mode reasoning, Gemini 3.1 Pro leads on GPQA Diamond at 94.3 percent and scores 77.1 percent on ARC-AGI-2, with native, highly reliable Google Search grounding for real time factual questions. A further update, Gemini 3.5 Pro, is expected imminently and could reshuffle this entire ranking on arrival

Best value frontier option, Grok 4.5 became publicly available this month and is currently the cheapest model inside the genuine top tier by GPQA Diamond score, strongest specifically for real time X and web context that other labs can't easily replicate

Best mid tier value pick, Qwen 3.7 Max from Alibaba, the highest ranked Chinese model on most current leaderboards, scoring competitively on math heavy benchmarks like HMMT at a fraction of frontier pricing

Open weights leaders, DeepSeek V4 Pro, GLM-5.2 from Z.AI, and Kimi K2.6 and K2.7 from Moonshot AI now trade blows with closed frontier models on reasoning and coding benchmarks at dramatically lower cost, DeepSeek's models in particular have been pushing frontier class quality below 1 dollar per million tokens with launch pricing

Market share, by actual web visit share this remains a very different picture from the benchmark rankings, ChatGPT holds roughly 54 percent of worldwide web traffic across major AI chatbots, Google Gemini sits at around 28 percent and has grown roughly 450 percent year over year, and Claude has grown the fastest of any major assistant, up roughly 855 percent year over year to about 9 percent worldwide share, driven heavily by developer adoption

The honest caveat that applies to every list like this, benchmark scores this close together, often within 2 to 3 percentage points, are functionally indistinguishable for most everyday use, the practical choice usually comes down to task fit, ecosystem integration and price rather than chasing whichever model tops a specific leaderboard that week according to AI Leaderboard 2026: Compare & Rank 300+ Top AI Models by Intelligence, Speed & Price

[EdgeRatedR86]

Claude Fable 5 being switched off and back on by an actual export control order is such a strange first for the industry, wonder how many more models end up with a regulatory on off switch like that going forward

[DiamondDallas]

The gap between benchmark leader and actual market share leader is the most interesting part of this whole list, Claude tops a lot of capability rankings but ChatGPT still dominates raw usage by a mile

[Janette_63]

Gemini 3.5 Pro being teased as imminent right as this list gets published is exactly the kind of timing that makes any ranking like this go stale within weeks

[PromptEcho]

Grok going from locked behind a 300 dollar tier to being one of the cheapest frontier options is a fast turnaround for that model specifically

[Jason99]

The open weight models closing the gap on reasoning and coding while costing a fraction of the price is probably the more consequential long term story buried in here than any single model's rank

[SilverRider]

Claude's 855 percent year over year growth is a wild number, that's not incremental growth, that's a legitimate shift in who developers are actually building on

[Lantern]

Appreciate the honest caveat at the end about benchmark differences being functionally meaningless at this point, way more useful advice than just handing over another top model crown

[Odd Voyager]

These rankings feel like trying to rank F1 cars mid-race.

By the time you finish the list, someone has already overtaken.

Still useful as a snapshot, just have to treat it as temporary by design

[BradBytheway]

One thing that stands out is how narrow the gaps are getting at the top.

It is less about one model dominating and more about slight differences in strengths.

Coding, reasoning, multimodal, each one edges ahead somewhere.

[StringTheory32]

Benchmark rankings are helpful, but real-world use tells a slightly different story.

Some models feel better in conversation, others shine in structured tasks.

Hard to capture that nuance in a single list

[LostReece55]

Feels like we are moving from "which is best" to "which is best for this task".

More like choosing tools than crowning a champion.

Probably a healthier way to think about it.

[Rapid Andrew]

There is also a bit of "what are you optimizing for" in these rankings.

Speed, cost, accuracy, creativity, they all matter depending on the use case.

No single winner across everything.

[Skibidi98]

At this point, any list is more of a conversation starter than a definitive ranking.

Still fun to compare though, and it highlights just how quickly things are improving

[WhatUQuant]

The Gemini timing is almost comedic.

New model drops right as the rankings go live, instantly outdated.

This space moves faster than the lists can keep up with

[TransformerCipher]

There is also the ecosystem factor.

APIs, tooling, integrations, documentation.

A slightly weaker model with better tooling can end up being more useful in practice

[KDB]

Interesting how smaller or more efficient models are creeping closer to the top tier.

Not necessarily beating them, but narrowing the gap enough to matter.

That could shift adoption patterns quite a bit.