AI Model Timeline 2026 — Major Releases Visualization

var models = [
{name:"Claude 4 Opus",vendor:"Anthropic",date:"2026-01-15″,dateStr:"Jan 15″,tags:[„reasoning","multimodal"],"params":"~900B (MoE)",mmlu:"92%",swebench:"92%",gpqa:"88%",notes:"1M context, extended thinking, SOTA safety",color:"#d97706″},
{name:"Gemini 2.5 Ultra",vendor:"Google DeepMind",date:"2026-02-03″,dateStr:"Feb 3″,tags:[„multimodal","reasoning","vision"],"params":"~1.8T (MoE)",mmlu:"93%",swebench:"89%",gpqa:"90%",notes:"Native vision/audio/text, deeply integrated in Google ecosystem",color:"#1a73e8″},
{name:"GPT-5″,vendor:"OpenAI",date:"2026-03-20″,dateStr:"Mar 20″,tags:[„reasoning","code","multimodal"],"params":"~5T (MoE, sparse)",mmlu:"94%",swebench:"93%",gpqa:"91%",notes:"Agentic capabilities, 40% API price drop vs GPT-4o",color:"#10a37f"},
{name:"Grok 3″,vendor:"xAI / X",date:"2026-03-01″,dateStr:"Mar 1″,tags:[„reasoning","code"],"params":"~2.5T",mmlu:"89%",swebench:"86%",gpqa:"83%",notes:"Trained on 100K H100s, strong real-time data from X",color:"#111″},
{name:"Qwen 3 Max",vendor:"Alibaba",date:"2026-04-08″,dateStr:"Apr 8″,tags:[„multimodal","opensource","reasoning"],"params":"~720B (MoE)",mmlu:"90%",swebench:"88%",gpqa:"85%",notes:"Open weight, strong Chinese/English bilingual, multimodal",color:"#ff6a00″},
{name:"Llama 4 Behemoth",vendor:"Meta",date:"2026-04-24″,dateStr:"Apr 24″,tags:[„opensource","multimodal"],"params":"~400B",mmlu:"90%",swebench:"87%",gpqa:"86%",notes:"Open weight license, 100+ language support, MoE",color:"#0078f0″},
{name:"Mistral Large 3″,vendor:"Mistral AI",date:"2026-04-01″,dateStr:"Apr 1″,tags:[„opensource","reasoning","code"],"params":"~240B",mmlu:"88%",swebench:"85%",gpqa:"84%",notes:"Open source, European champion, strong efficiency",color:"#ff6900″},
{name:"Llama 4 Scout",vendor:"Meta",date:"2026-04-24″,dateStr:"Apr 24″,tags:[„opensource","code"],"params":"~109B total, 17B active (MoE)",mmlu:"86%",swebench:"82%",gpqa:"80%",notes:"Runs on single GPU, edge deployment",color:"#0078f0″},
{name:"Claude 4 Sonnet",vendor:"Anthropic",date:"2026-05-01″,dateStr:"May 1″,tags:[„reasoning","multimodal"],"params":"~200B",mmlu:"90%",swebench:"90%",gpqa:"87%",notes:"Cost-efficient, faster than Opus, same safety",color:"#d97706″},
{name:"DeepSeek R2″,vendor:"DeepSeek",date:"2026-05-18″,dateStr:"May 18″,tags:[„opensource","reasoning","code","multimodal"],"params":"~671B (MoE, 37B active)",mmlu:"93%",swebench:"91%",gpqa:"90%",notes:"Fully open source, 1/20th training cost of GPT-5″,color:"#1e40af"},
{name:"ERNIE 5.0″,vendor:"Baidu",date:"2026-05-01″,dateStr:"May 1″,tags:[„multimodal","vision"],"params":"~1.2T",mmlu:"87%",swebench:"84%",gpqa:"82%",notes:"Leading Chinese multimodal model, 50M+ enterprise users",color:"#2563eb"},
{name:"Command R+",vendor:"Cohere",date:"2026-01-30″,dateStr:"Jan 30″,tags:[„reasoning","code"],"params":"~104B",mmlu:"85%",swebench:"80%",gpqa:"78%",notes:"RAG-optimized, 128K context, multilingual",color:"#059669″},
{name:"GPT-5 Mini",vendor:"OpenAI",date:"2026-03-20″,dateStr:"Mar 20″,tags:[„code"],"params":"~50B",mmlu:"88%",swebench:"87%",gpqa:"84%",notes:"Cost-optimized for API consumers and edge",color:"#10a37f"},
{name:"Kimi k1.5″,vendor:"Moonshot AI",date:"2026-01-20″,dateStr:"Jan 20″,tags:[„reasoning","multimodal"],"params":"~300B",mmlu:"89%",swebench:"85%",gpqa:"84%",notes:"Long context (2M tokens), strong Chinese market",color:"#7c3aed"}
];
function $(s){return document.querySelector(s)}
function render(){
var vendorFilter = $('filterVendor').value;
var tagFilter = $('filterTag').value;
var sortBy = $('sortBy').value;
var filtered = models.filter(m => {
if (vendorFilter !== 'all' && m.vendor !== vendorFilter) return false;
if (tagFilter !== 'all' && !m.tags.includes(tagFilter)) return false;
return true;
});
var html = '';
filtered.forEach((m,i) => {
var tagsHtml = m.tags.map(t => ''+t+'').join('');
});
$(‚timeline').innerHTML = ‚
// Update summary
$(‚totalModels').textContent = filtered.length;
var vendors = [...new Set(filtered.map(m=>m.vendor))];
$(‚vendors').textContent = vendors.length;
var totalP = filtered.reduce((s,m) => s + parseParams(m.params), 0);
$(‚avgParams').textContent = formatParams(totalP / filtered.length);
function parseParams(str) {
var m = str.match(/~?([d.]+)s*(T|B)/i);
if (!m) return 0;
var v = parseFloat(m[1]);
return m[2].toUpperCase() === 'T' ? v * 1000 : v; // in B
}
function formatParams(billions) {
if (billions >= 1000) return (billions/1000).toFixed(1)+'T';
if (billions >= 1) return billions.toFixed(0)+'B';
return (billions*1000).toFixed(0)+'M';
}
// Populate vendor filter
var vendors = [...new Set(models.map(m=>m.vendor))].sort();
var vendorSelect = $('filterVendor');
vendors.forEach(v => {
var opt = document.createElement('option');
opt.value = v; opt.textContent = v;
});

Major AI model releases in 2026 H1 — filterable, sortable, interactive.

All Vendors

All Capabilities
Multimodal
Reasoning
Open Source
Code
Vision+

Sort by Date
Sort by Parameters
Sort by MMLU-Pro

Benchmark scores from official vendor disclosures and independent evaluations. Parameter counts are estimates for some models. Last updated: June 2026.