AI Evolution

Frontier & Tier 1 Models — November 2023 to October 2026

Last updated: 2026-10-01

November 2023
GPT-4 Turbo Launch
OpenAI
OpenAI introduces GPT-4 Turbo with a 128K context window, improved function calling, and lower costs. Marks a significant evolution in the multimodal frontier.
128K context Function calling Cost reduction
December 2023
Gemini Launch
Google DeepMind
Google releases Gemini in Ultra, Pro, and Nano variants, demonstrating multimodal capabilities across text, image, video, and audio. Establishes Google as frontier competitor.
Multimodal 3 variants Video capable
January 2024
Claude 3 Family
Anthropic
Anthropic releases Claude 3 (Opus, Sonnet, Haiku) with improved reasoning, vision capabilities, and longer context windows. Establishes a three-tier competitive lineup.
Vision Long context 3 tiers
February 2024
Llama 3 Preview
Meta
Meta previews Llama 3, signaling major improvements in open-source foundation models. Focus on instruction-following and multimodal capabilities.
Open source Foundation model Multimodal
April 2024
Grok-1 Release
xAI
xAI releases Grok-1 as an open-weights model, emphasizing real-time knowledge and powerful reasoning capabilities for frontier-class performance.
Open weights Real-time data Reasoning
May 2024
GPT-4o Launch
OpenAI
OpenAI releases GPT-4o (omni), the first natively multimodal model handling text, image, and audio in a unified architecture. Sets new frontier benchmarks.
Native multimodal Audio input/output Flagship model
June 2024
Claude 3.5 Sonnet
Anthropic
Anthropic releases Claude 3.5 Sonnet with improved coding, mathematical reasoning, and extended context. Bridges Sonnet and Opus capability gap.
Better coding Extended context Cost optimized
September 2024
Llama 3.1 Full Release
Meta
Meta releases Llama 3.1 (8B, 70B, 405B) as fully open-source with extended context and instruction-following improvements. Strengthens open-source frontier.
405B variant Open source 3 sizes
October 2024
GPT-4 Turbo 2024
OpenAI
OpenAI updates GPT-4 Turbo with improved reasoning, code generation, and knowledge cutoff refresh. Maintains frontier leadership while preparing for GPT-5.
Improved reasoning Latest knowledge Stable flagship
December 2024
Gemini 2.0 Launch
Google DeepMind
Google releases Gemini 2.0 with enhanced reasoning, longer context windows, and improved code generation capabilities across all model tiers.
Extended context Better reasoning Code synthesis
January 2025
Claude 4 (Opus 4 / Sonnet 4)
Anthropic
Anthropic ships Claude Opus 4 and Sonnet 4 as the fourth-generation line, with stronger coding, extended context, and agent workflows across API and Claude products.
Opus 4 Sonnet 4 Agents
March 2025
GPT-5 Family
OpenAI
OpenAI rolls out the GPT-5 model family across API and ChatGPT as its next-generation reasoning and multimodal stack, replacing GPT-4o as the primary frontier line.
GPT-5 Reasoning API + ChatGPT
June 2026
Claude Fable 5 and Mythos 5
Anthropic
Anthropic releases Claude Fable 5 widely and Claude Mythos 5 as invitation-only (Project Glasswing). Access interrupted 12 Jun–1 Jul.
Frontier GA Gated (Mythos)
June 2026
North Mini Code
Cohere
Cohere ships the first North-family 30B-A3B Apache coding MoE, extending the stack into agentic coding.
Shipped Open weights New family
June 2026
Claude Sonnet 5
Anthropic
Anthropic releases Claude Sonnet 5 as the most agentic Sonnet tier with 1M context for production agent workloads.
GA Agents API
June 2026
GLM-5.2
Z.ai
Z.ai releases GLM-5.2 with 1M context and MIT open weights for agentic and long-horizon coding.
Shipped Open weights 1M context
July 2026
GPT-5.6 Family GA
OpenAI
OpenAI ships GPT-5.6 Sol / Terra / Luna in API and ChatGPT as the current frontier family.
GA Frontier API
July 2026
Gemini 3.6 Flash + 3.5 Flash-Lite
Google
Google GA's Gemini 3.6 Flash and 3.5 Flash-Lite as flash workhorse tiers. Gemini 4 remains pre-training only, not a product.
GA Agents Not Gemini 4
July 2026
Hy3 GA
Tencent
Tencent Hy3 reaches GA (295B-A21B, Apache 2.0) after April preview, becoming the current language flagship.
GA Open weights Agent
July 2026
Kimi K3
Moonshot
Moonshot ships Kimi K3 (2.8T native-multimodal, 1M context); open weights follow 27 Jul.
Shipped 1M context Open weights
July 2026
Claude Opus 5
Anthropic
Anthropic releases Claude Opus 5 as the recommended default for complex agentic coding.
GA Frontier Coding
July 2026
Grok 4.5
SpaceXAI
SpaceXAI ships Grok 4.5 as the then-current coding and agents flagship on x.ai.
GA Frontier Coding
July 2026
MiniMax H3
MiniMax
MiniMax ships H3 omni video with stereo audio on API (31 Jul), up to 15s / 2K output.
Shipped Video Omni
August 2026
Qwen3.8-Max + Open Weights
Alibaba
Alibaba GA's hosted Qwen3.8-Max (3 Aug) and releases 2.4T-A95B open weights, followed by 27B Apache VL.
GA Open weights Flagship
August 2026
Gemini 3.7 Flash
Google
Google GA's Gemini 3.7 Flash as the current public workhorse for coding and agents.
GA Flash Coding
August 2026
Grok 4.6
SpaceXAI
SpaceXAI ships Grok 4.6 as the current Grok flagship with 500k context for text, code, and agents.
GA Frontier Agents
August 2026
DeepSeek V4-Pro GA
DeepSeek
DeepSeek GA's V4-Pro-0813 on App, Web, and API; deepseek-chat / deepseek-reasoner aliases already retired in July.
GA Agents API
August 2026
GLM-5.3 Hosted
Z.ai
Z.ai GA's hosted GLM-5.3 on the same base as 5.2 with post-training only; full open weights follow 28 Aug under custom glm-5.3 license.
Hosted GA Text-only Weights follow
August 2026
Nemotron 3.5 Lightning
NVIDIA
NVIDIA releases Nemotron 3.5 Lightning (30B-A3B open MoE) for agent workhorses via Hugging Face and NIM.
Shipped Open weights NIM
August 2026
Muse Spark 1.2 + Glimmer 30B
Meta MSL
Meta MSL ships hosted Muse Spark 1.2 as default and Apache Muse Glimmer 30B for local agentic workloads.
API Open weights Agents
August 2026
Mistral Regional Inference + Hosted GLM-5.2
Mistral
Mistral GA's EU/US regional endpoints and adds hosted third-party open models starting with Z.ai GLM-5.2 preview.
GA endpoints Third-party preview EU
August 2026
Gemini 3.5 Transcribe
Google
Google GA's Gemini 3.5 Transcribe (changelog 26 Aug) as a Chirp 3 successor for speech-to-text via Gemini API, AI Studio, and Gemini Enterprise Agent Platform.
Frontier GA Speech
August 2026
Gemini Omni Flash GA
Google
Google GA's Gemini Omni Flash (gemini-omni-1.1-flash, changelog 27 Aug); gemini-omni-flash-preview deprecate date 30 Sep past — use GA ID.
Frontier GA Video
August 2026
Qwen3.8-Flash-Next
Alibaba
Alibaba releases Qwen3.8-Flash-Next open weights as an architecture preview toward Qwen4 — not Qwen4 itself; hosted Flash API not on Model Studio.
T1 Open weights Not Qwen4
August 2026
GLM-5.3-Flash
Z.ai
Z.ai ships GLM-5.3-Flash as the first natively multimodal GLM-5 tier with MIT Flash weights; full GLM-5.3 weights release separately 28 Aug.
T2 Shipped Open weights (Flash only)
August 2026
Granite 4.2
IBM
IBM releases Granite 4.2 as an Apache dense reasoning family with native thinking plus Speech 5.0 Turbo CTC for high-throughput ASR.
New T2 Open weights Apache
August 2026
GLM-5.3 Full Weights
Z.ai
Z.ai releases ungated GLM-5.3 full weights on Hugging Face under custom glm-5.3 license (not MIT).
T2 Open weights Custom license
August 2026
Qwen3.8-Flash Hosted
Alibaba
Alibaba adds hosted qwen3.8-flash to Model Studio as catalog-live flash tier; Flash-Next open weights remain architecture preview, not Qwen4.
T1 Hosted Not Qwen4
August 2026
Hy4 Preview
Tencent
Tencent releases Hy4 preview as Apache open weights on Hugging Face; not GA. Hy3 remains the language flagship.
T2 Preview Open weights
September 2026
Claude Fable 5.1 / Mythos 5.1
Anthropic
Anthropic GA's Claude Fable 5.1 and ships Claude Mythos 5.1 as invitation-only (Project Glasswing); not GA.
Frontier GA (Fable) Gated (Mythos)
September 2026
Gemini 3.8 Flash
Google
Google GA's Gemini 3.8 Flash as the current public workhorse; 3.8 Flash Cyber is Fairwind-gated, not public GA. Not Gemini 4.
Frontier GA Not Gemini 4
September 2026
Muse Spark 1.3
Meta
Meta MSL rolls out Muse Spark 1.3 in Muse Code and Meta Model API as the recommended agentic/coding tier; max / xhigh reasoning is live.
T1 API Agents
September 2026
Muse Voice Transcribe
Meta
Meta MSL ships Muse Voice Transcribe 1.0 as a speech-to-text family; separate from Spark, not a replacement.
T1 Shipped Speech
September 2026
GPT-6 Astra
OpenAI
Ships GPT-6 Astra (gpt-6-astra) as current API/ChatGPT flagship; first Critical cyber under Preparedness Framework; Daybreak for advanced cyber.
Frontier GA Critical cyber
September 2026
GPT-Image-2.5 Flare / Sunburst
OpenAI
New image API line (8 Sep): Flare = fast default; Sunburst = precision editing.
Frontier GA Image
September 2026
GPT-Live-1
OpenAI
Full-duplex voice model in the API (10 Sep); pairs with backend models.
Frontier GA Voice
September 2026
Lyria 3.5 GA (status fix)
Google
Gemini API changelog lists Lyria 3.5 (lyria-3.5) as GA (3 Sep); music generation on Gemini API.
Frontier GA Music
September 2026
DeepSeek V4.1-Flash
DeepSeek
New architecture Flash tier with native vision (10 Sep; deepseek-flash); current default. No V4.1-Pro.
T1 GA Multimodal
September 2026
Grok 4.7
SpaceXAI
Flagship coding/agents/knowledge model GA (~21 Sep); API ID grok-4.7; 500k context; same $2/$6 list class as 4.6.
Frontier GA Coding
September 2026
GPT-6 Sol + GPT-6 Luna
OpenAI
Mid + efficient GPT-6 tiers GA (22 Sep); Astra remains flagship. IDs gpt-6-sol / gpt-6-luna.
Frontier GA GPT-6
September 2026
Claude Opus 5.5
Anthropic
New Opus default GA (22 Sep); claude-opus-5-5; $4/$20; Fable-class performance claim at lower cost.
Frontier GA Agents
September 2026
Gemini 3.8 Flash TTS + Flash-Lite TTS
Google
Next-gen TTS GA (22 Sep) + Voices endpoint; IDs gemini-3.8-flash-tts / gemini-3.8-flash-lite-tts.
Frontier GA Voice
September 2026
Perplexity Sonar EOL (completed)
Perplexity
Sonar Chat Completions support ended 27 Sep; migrate/reformulate to Agent API presets; sync+stream may still work as reformulations; async Sonar unsupported.
T2 EOL Migrate
September 2026
DeepSeek V4-Pro continuity (ops fix)
DeepSeek
Pro API (deepseek-v4-pro / V4-Pro-0813) continues after 14 Sep with unchanged billing per ops docs/changelog — supersedes Pro→Flash routing on older news copy; no V4.1-Pro.
T1 Docs fix Not V4.1-Pro
September 2026
Agents API (public beta)
OpenAI
Managed Codex harness for durable agent sessions, tools, MCP, and sandboxes; US data residency only; no ZDR in beta. Pairs with gpt-6-astra et al.
Frontier Public beta Agents
September 2026
Gemini 3.8 Live + Live Extended Thinking
Google
Audio-to-audio Live API models GA (15 Sep); low-latency (gemini-3.8-live) vs high-reasoning live variants (gemini-3.8-live-extended-thinking).
Frontier GA Voice
September 2026
TypeSafe exits stealth with System One / Jev
TypeSafe AI
After ~2 years in stealth, TypeSafe opens early access to Jev, its first System One model — typed calibrated decisions for software (not a chat LLM), trained with RLCD. Announces ~$40M seed led by DCVC.
New model class Early access $40M seed
September 2026
Claude Sonnet 5.5
Anthropic
Mid-tier Claude 5.5 GA (28 Sep); API ID claude-sonnet-5-5; $2/$10; speed/intelligence combo; Haiku 5.5 still roadmap.
Frontier GA Coding
September 2026
GPT-6.1 Sol
OpenAI
Mid-tier GPT-6.1 GA (29 Sep); ID gpt-6.1-sol; near-Astra coding/computer-use at $2/$10; Multi-agent beta.
Frontier GA GPT-6
September 2026
Agents API computer use
OpenAI
Public-beta Agents API gains OpenAI-hosted browser computer use (29 Sep); still beta (agents=v1).
Frontier Beta Agents
September 2026
Omni Flash preview EOL
Google
gemini-omni-flash-preview deprecate date 30 Sep past; use GA gemini-omni-1.1-flash.
Frontier EOL Video
October 2026
Antigravity preview-05 shutdown
Google
antigravity-preview-05-2026 shuts 5 Oct; migrate to antigravity-preview-09-2026.
Frontier Deadline Agents