Saturday, 27 June 2026 NewsletterSign inSubscribe
The Global AI Brief

The definitive ranking · 2026 edition

The Top 50 AI Foundational Models

The models enterprises are actually evaluating, scored on capability, openness, and what it takes to run them in production. Tap any model to open it.

Edited by the Global AI Forum research desk

The foundation-model field now ships a meaningful release almost every week, and the frontier moved twice in the last fortnight alone. Anthropic's Mythos-tier Fable 5 sits at the top of raw capability while its access stays restricted, and on 26 June OpenAI opened a limited preview of GPT-5.6 Sol. Most enterprise buyers have not caught up, and the thing that separates a shortlist from a deployment has shifted from capability to whether a model can be governed, costed, and run where the data lives.

We rank on three things a buyer cares about: capability, openness, and readiness. The score is the Global AI Brief composite. Availability is flagged in the access column rather than buried in the number, so a capability leader you cannot yet use is visible as exactly that. The field moves weekly, so treat the figures as a June 2026 snapshot.

Featured
Analysis

The two releases that reset the frontier this month

Charted

The open-weight surge, in four charts

Shortlist

Which models actually ship in the enterprise

The ranking

June 2026 snapshot · tap a row to expand
RankModelMakerReleasedContextAccessScore
1 Claude Fable 5 NewAnthropic · Mythos tier AnthropicJun 2026— Restricted 98
Claude Fable 5
Anthropic's Mythos-tier model and the capability leader on several public leaderboards. Access is restricted under US export controls, so for most buyers it reads as a frontier benchmark rather than a deployment option right now.
Released Jun 2026Access RestrictedTier MythosMaker Anthropic
98/100
GAI Brief composite
GPT-5.6 Sol
OpenAI's new flagship, the top of the GPT-5.6 series alongside Terra and Luna. Previewed on 26 June to a small group of trusted partners coordinated with the US government, with broad availability promised in the coming weeks.
Released Jun 2026Context 1.5MAccess PreviewMaker OpenAI
96/100
GAI Brief composite
Claude Opus 4.8
Anthropic's generally available frontier model. Tops the Artificial Analysis Intelligence Index and leads SWE-bench among shipping models, with the strongest agentic and computer-use scores in the set.
Released May 2026Context 200KAccess ClosedMaker Anthropic
95/100
GAI Brief composite
GPT-5.5
OpenAI's prior flagship and still a production default. Strong agentic coding, a one-million-token API window, and the broadest ecosystem of any model in the set.
Released Apr 2026Context 1MAccess ClosedMaker OpenAI
93/100
GAI Brief composite
Gemini 3.1 Pro
Google DeepMind's frontier model. Long-context and multimodal strengths, and a consistent leader on human-preference coding-arena play.
Released May 2026Context 1MAccess ClosedMaker Google DeepMind
92/100
GAI Brief composite
Claude Sonnet 4.6
Anthropic's value tier. Near-Opus quality at a lower price, with a one-million-token context window in beta, and a common default for high-volume work.
Released Apr 2026Context 1MAccess ClosedMaker Anthropic
89/100
GAI Brief composite
Grok 4.3
xAI's frontier model. Strong reasoning and a large context window, closed and served through the xAI and partner APIs.
Released Apr 2026Context 256KAccess ClosedMaker xAI
88/100
GAI Brief composite
GLM-5.2
Z.ai's flagship and the strongest open-weight model on several benchmarks. MIT-licensed, with a usable one-million-token context window, the largest of any current open model.
Released Jun 2026Context 1MAccess Open-weightLicence MIT
87/100
GAI Brief composite
DeepSeek V4 Pro
DeepSeek's open-weight frontier model and a value leader. Widely self-hosted, and a reference point for what a dollar of inference now buys.
Released May 2026Context 128KAccess Open-weightMaker DeepSeek
86/100
GAI Brief composite
Qwen 3.7 Max
Alibaba's proprietary flagship, and the cheapest model in the top tier per million tokens. Its open Qwen siblings anchor much of the open-weight ecosystem.
Released May 2026Context 256KAccess ClosedMaker Alibaba
84/100
GAI Brief composite
Kimi K2.7 Code
Moonshot AI's trillion-parameter open-weight model, tuned for long-horizon coding and noted for using markedly fewer thinking tokens than its predecessor.
Released Jun 2026Context 256KAccess Open-weightMaker Moonshot AI
83/100
GAI Brief composite
MiniMax M3
A multimodal open-weight model with a one-million-token window, released June, with strong SWE-bench results at a price that helped reset the floor for agents.
Released Jun 2026Context 1MAccess Open-weightMaker MiniMax
82/100
GAI Brief composite
Llama 4 Scout
Meta's open-weight model, most notable for a ten-million-token context window, the largest in wide use, and a default base for fine-tuning.
Released Mar 2026Context 10MAccess Open-weightMaker Meta
80/100
GAI Brief composite
Nemotron 3 Ultra
NVIDIA's 550-billion-parameter model under a fully permissive licence, aimed at teams that want frontier-adjacent quality they can own outright.
Released Jun 2026Context 128KAccess OpenMaker NVIDIA
78/100
GAI Brief composite
Devstral 2
Mistral AI's open-weight coding specialist, a 123-billion-parameter dense model built for agentic software work and portable deployment.
Released May 2026Context 256KAccess Open-weightMaker Mistral AI
76/100
GAI Brief composite
DeepSeek V4 Flash
DeepSeek's low-cost open-weight model that reset the price floor for agentic workloads, popular where cost-per-task is the deciding factor.
Released Apr 2026Context 128KAccess Open-weightMaker DeepSeek
74/100
GAI Brief composite
GPT-5.4
OpenAI's March flagship and a one-million-token predecessor still in wide production use, a useful baseline for migration decisions.
Released Mar 2026Context 1MAccess ClosedMaker OpenAI
73/100
GAI Brief composite
Qwen3 Coder 480B
Alibaba's open-weight coding model and the base for its proprietary Coder Plus, a workhorse for autonomous programming via tool calling.
Released 2026Context 256KAccess Open-weightMaker Alibaba
72/100
GAI Brief composite
Mistral Large 3
Mistral's closed-API flagship, widely used in European enterprise deployments that need GDPR-native hosting options.
Released 2026Context 128KAccess ClosedMaker Mistral AI
71/100
GAI Brief composite
Command R+ Ultra
Cohere's retrieval-optimised model, the default choice when RAG and enterprise search are the primary workload.
Released 2026Context 128KAccess ClosedMaker Cohere
70/100
GAI Brief composite
Gemini 3.0 Flash
Google's cost-efficient tier, fast inference at a fraction of Pro pricing, and a one-million-token window.
Released Apr 2026Context 1MAccess ClosedMaker Google DeepMind
69/100
GAI Brief composite
Claude Haiku 4.5
Anthropic's lowest-cost and fastest model, preferred for high-throughput customer-facing tasks.
Released 2026Context 200KAccess ClosedMaker Anthropic
68/100
GAI Brief composite
Phi-4 Ultra
Microsoft's small-model leader, optimised for on-device and edge deployment with strong reasoning relative to size.
Released 2026Context 128KAccess Open-WeightMaker Microsoft
67/100
GAI Brief composite
Gemma 3 27B
Google's open-licence compact model, a common base for fine-tuning in constrained environments.
Released 2026Context 128KAccess OpenMaker Google DeepMind
66/100
GAI Brief composite
Llama 4 Maverick
Meta's full-capability open-weight model, complementing Scout's long-context edge with stronger instruction following.
Released 2026Context 1MAccess Open-WeightMaker Meta
65/100
GAI Brief composite
Yi-Lightning 2
01.AI's frontier model, competitive on Chinese-language enterprise tasks and growing in international deployments.
Released 2026Context 200KAccess ClosedMaker 01.AI
64/100
GAI Brief composite
Falcon 3 180B
The Technology Innovation Institute's fully open flagship, notable for permissive licensing and sovereign deployment.
Released 2026Context 64KAccess OpenMaker TII
63/100
GAI Brief composite
AYA Expanse 35B
Cohere's multilingual model, covering over 23 languages and suited to international enterprise workloads.
Released 2026Context 128KAccess Open-WeightMaker Cohere
62/100
GAI Brief composite
InternLM 3 70B
Shanghai AI Lab's open model, strong on STEM benchmarks and deployed widely in Chinese research institutions.
Released 2026Context 200KAccess OpenMaker Shanghai AI Lab
61/100
GAI Brief composite
Baichuan 3 Turbo
Baichuan's enterprise-focused closed model, a default for Chinese-market deployments requiring local data residency.
Released 2026Context 128KAccess ClosedMaker Baichuan AI
60/100
GAI Brief composite
ERNIE 5.0
Baidu's conversational flagship, embedded across its own ecosystem and licensed to third-party platforms.
Released 2026Context 128KAccess ClosedMaker Baidu
59/100
GAI Brief composite
Spark V4
iFlytek's enterprise model, strongest in voice-adjacent and education vertical deployments.
Released 2026Context 64KAccess ClosedMaker iFlytek
58/100
GAI Brief composite
Skywork-o3
Kunlun Tech's reasoning-tuned open-weight model, gaining traction in mathematics and code competition benchmarks.
Released 2026Context 128KAccess Open-WeightMaker Kunlun Tech
57/100
GAI Brief composite
Granite 3.3 8B
IBM's open enterprise model, designed for low-latency inference on-premise and in regulated financial environments.
Released 2026Context 128KAccess OpenMaker IBM
56/100
GAI Brief composite
Orca 3 13B
Microsoft Research's small reasoning model, a teaching-via-explanation approach that punches above its parameter count.
Released 2026Context 64KAccess OpenMaker Microsoft Research
55/100
GAI Brief composite
Mixtral 8x22B v2
Mistral's mixture-of-experts open flagship, the compute-efficient option when the serving infrastructure is constrained.
Released 2026Context 64KAccess OpenMaker Mistral AI
54/100
GAI Brief composite
SeaLLM 4
Alibaba DAMO's Southeast Asia-focused multilingual model, covering Thai, Vietnamese, Malay, and others.
Released 2026Context 128KAccess OpenMaker DAMO Academy
53/100
GAI Brief composite
OpenHermes 4
Nous Research's community-tuned model, a widely forked base for instruction-following fine-tunes.
Released 2026Context 128KAccess OpenMaker Nous Research
52/100
GAI Brief composite
Jamba 2.5
AI21's hybrid SSM-Transformer architecture, pitched at long-document analysis and knowledge retrieval at enterprise scale.
Released 2026Context 256KAccess ClosedMaker AI21 Labs
51/100
GAI Brief composite
OLMo 2 7B
Allen AI's fully open model with released training data and code, valued for AI safety and reproducibility research.
Released 2026Context 128KAccess OpenMaker Allen AI
50/100
GAI Brief composite
Titan Text Premier
Amazon's managed text model on Bedrock, chosen when the AWS infrastructure constraint makes managed API the default.
Released 2026Context 32KAccess ClosedMaker Amazon
49/100
GAI Brief composite
PaLM-E 2
Google's embodied multimodal model, used in robotics and physical-world planning research.
Released 2026Context 128KAccess ClosedMaker Google DeepMind
48/100
GAI Brief composite
Hyper-Clova X 2
Naver's Korean-language enterprise model, the default for Korea-market deployments requiring local compliance.
Released 2026Context 32KAccess ClosedMaker Naver
47/100
GAI Brief composite
Llama 3.3 70B
Meta's prior-generation compact open-weight model, still widely in production and a stable fine-tuning base.
Released 2025Context 128KAccess Open-WeightMaker Meta
46/100
GAI Brief composite
Nemotron 2 Ultra
NVIDIA's prior open model generation, still deployed at scale for on-prem AI factory workloads.
Released 2025Context 128KAccess OpenMaker NVIDIA
45/100
GAI Brief composite
WizardLM 2 13B
A community fine-tune focused on instruction complexity, popular in agentic research pipelines.
Released 2025Context 64KAccess OpenMaker WizardLM
44/100
GAI Brief composite
ChatGLM 6B
The compact open-weight Chinese model from Tsinghua KEG Lab, the entry point for Chinese-language fine-tuning.
Released 2025Context 32KAccess OpenMaker Tsinghua/Zhipu
43/100
GAI Brief composite
Mistral Nemo 2
Mistral's small multilingual open model, designed for EU deployment with a 128K context at four billion parameters.
Released 2025Context 128KAccess OpenMaker Mistral AI
42/100
GAI Brief composite
Phi-3.5 Mini
Microsoft's edge-optimised SLM, the benchmark reference for small models under four billion parameters.
Released 2025Context 128KAccess OpenMaker Microsoft
41/100
GAI Brief composite
StableLM 3 3B
Stability AI's compact open model, used for embedded and constrained-environment deployments.
Released 2025Context 32KAccess OpenMaker Stability AI
40/100
GAI Brief composite

All 50 models · June 2026 snapshot · tap any row to expand

How we score

Every model carries one composite out of 100, built from three weighted inputs. Capability sets the ranking, openness and readiness adjust it, and availability is flagged in the access column rather than hidden in the number.

50
Capability
Reasoning, coding, and agentic benchmarks, weighted toward enterprise tasks.
20
Openness
Weights, licence, and how freely the model can be self-hosted.
30
Readiness
Governance tooling, deployment options, and cost to run in production.

Model names, makers, and access reflect the public record as of June 2026. The composite scores are the Global AI Brief' own, shown here as a working snapshot, and the field moves weekly.

More rankings from The Global AI Brief
50
Ranking
Top 50 AI Agents in Production
The autonomous systems actually running inside enterprises today.
Open the list → 50
Ranking
Top 50 Open-Weight Models
The models you can self-host, ranked by capability and licence.
Open the list → 50
Ranking
Top 50 Generative Media Tools
Image, video, and audio systems, ranked for production use.
Open the list →

Don't just keep up. Get ahead.

The enterprise AI brief, handpicked for you each morning. Free.