AI model release timeline
Every model we cover, by the date it came out, newest first. Useful when someone says "the new Gemini" and you want to know which one they mean, or when you need to check whether a model you read about last month has already been replaced.
November 2026 · 3 releases
8 Flash. It powers Google Search for all users and is in the Gemini app, the Gemini API and Google Antigravity.
2 (9 November 2026) is Anthropic's top public model for long, hard technical work: large codebases, maths, robotics and research that runs for hours.
6 (4 November 2026) is OpenAI's image model for generation and editing, released in two versions: Flare for fast iteration and Sunburst for detailed visual work.
October 2026 · 7 releases
Gemini 4 Carbon (announced 28 October 2026, generally available 10 November) is what Google calls its strongest model, with state-of-the-art cybersecurity and…
Genie 4 (announced 27 October 2026) is Google DeepMind's world model: you move through an environment it generates in real time, with the keyboard, and steer it by…
1 Terra (26 October 2026) is OpenAI's fast, lower-cost model for computer use, visual reasoning and writing — filling forms, drafting proposals and working through apps.
1 Pro (23 October 2026) is Moonshot AI's flagship, a "3T-class" model with a 1M-token context, built for coding, 3D modelling and financial research. 4%.
Muse Spark Open (15 October 2026) is Meta's open-weight release of its Muse Spark model — a mixture-of-experts with about 816 billion total and 42 billion active…
Gemini 4 Argon is Google's frontier model, generally available since 13 October 2026 after a first release limited to cyber defenders.
5 (7 October 2026) is Anthropic's small, fast model — which Anthropic calls the cheapest, fastest and most capable small model it has released. 5.
September 2026 · 9 releases
5 — aimed at well-scoped everyday work, bug fixing and polished documents, slides and spreadsheets.
5 family and its new lead model for agentic coding, computer use and knowledge work. 1 on most work while costing 40% less to run than Opus 5.
Gemini 3.8 Live (15 September 2026) is Google's live dialogue model for voice agents — built for near real-time conversation rather than turn-by-turn chat, with visual…
8 Live. It adds multi-step reasoning to real-time conversation, and can explain its thinking while it works rather than going silent.
DeepSeek V4.1 Flash (10 September 2026) is DeepSeek's current main model — a 552B-parameter mixture-of-experts with 8B active parameters for input and 16B for output, a…
GPT-6 Astra (3 September 2026) is OpenAI's flagship and the first of the GPT-6 line — built for end-to-end computer use, browser work, coding and research, with a…
75 per million input tokens, roughly a thirteenth of what the current flagships charge. That rate is introductory and expires on 31 December 2026.
25 per million, which is what actually lowers the bill on agentic work.
1 — identical weights, identical pricing — running with more permissive safeguards in cybersecurity and the life sciences. It is invitation-only, not a consumer product.
August 2026 · 4 releases
Seedance 2.5 (August 2026) is ByteDance's update to its leaderboard-topping Seedance line — sharper detail and longer coherent clips than Seedance 2.0, still reached…
Qwen 4 (August 2026) is Alibaba's latest open-weight model family, spanning small on-device sizes up to a large flagship — strong at coding and math, with native…
Hailuo 3 (August 2026) is MiniMax's latest video model — known for expressive character motion and camera work, at a lower price than Veo or Sora.
Mistral Large 3 (August 2026) is the French lab's flagship — a strong multilingual model with open weights for the base version and a hosted API for the full model,…
July 2026 · 5 releases
Claude Opus 5 (24 July 2026) is Anthropic's thoughtful, proactive daily-driver model — it comes close to the frontier intelligence of Claude Fable 5 at HALF the price,…
8T-parameter model with native vision and a 1M-token context, built for long-horizon coding, knowledge work and reasoning.
5-class model — running at ~350 output tokens/sec with a strong price-to-performance ratio, and now with built-in computer use.
5 Flash while using ~17% fewer output tokens (up to 65% on some coding benchmarks), at a lower cost per token. Built for scaling AI agents.
5 (July 2026) is xAI's smartest model, built for coding, agentic tasks and knowledge work and trained alongside Cursor.
June 2026 · 9 releases
Claude Sonnet 5 (June 2026) is Anthropic's most agentic Sonnet yet — strong planning, browser/terminal tool use and long autonomous runs at a mid-tier price.
Gemini Omni Flash (June 2026) is Google's cost-efficient video generation + conversational editing model — create and edit video from text/image/video inputs…
6T-parameter open MoE (~48B active) built for agentic coding with a native 1M context — notable for being trained and served entirely on Chinese-made chips, under MIT.
Nano Banana 2 Lite (June 2026) is Google's fastest, most cost-efficient image generation/editing model — it can generate an image in as little as ~4 seconds.
GPT-5.6 (previewed June 2026) is OpenAI's new three-tier frontier family — flagship Sol, balanced Terra and low-cost Luna — with a new maximum reasoning effort and an…
1 Pro (June 2026) is ByteDance's enterprise agent-focused flagship — strong on coding, desktop/mobile agents and multimodal/video understanding.
GLM-5.2 (June 2026) is Z.AI's strongest open-weight model for long-horizon autonomous coding — a 744B MoE with a stable 1M-token context, released under MIT.
7 Code (June 2026) is Moonshot AI's coding-focused agentic model, tuned for long-horizon software-engineering workflows, with a faster 'Highspeed' variant.
8. Adaptive thinking is always on, and it uses a 1M-token context with vision and full tool use.
May 2026 · 2 releases
8 (May 2026) is Anthropic's top tier — best-in-class agentic coding and reasoning, a 1M-token context and careful, low-error output, priced and paced as a flagship.
Gemini 3.5 Flash (May 2026) is Google's fast tier that punches at flagship level on coding and agents at a fraction of Pro's price — now the default Gemini app / Search…
April 2026 · 5 releases
Grok 4.3 (April 2026) is xAI's flagship — a 1M-context, always-reasoning multimodal model with configurable 'effort', priced unusually low for a top-tier model.
Mistral Medium 3.5 is a 128B dense all-rounder that fuses instruct, reasoning and coding into one self-hostable model aimed at agentic workflows.
DeepSeek V4-Flash is the cheap, fast 1M-context MoE — extraordinary tokens-per-dollar that makes it the budget default for high-volume Indian deployments.
6T/49B-active MoE with a 1M context and built-in reasoning that subsumes the R-series.
GPT-5.5 (April 2026) is OpenAI's flagship — best-in-class agentic coding, frontier reasoning and a ~1M-token context, powering the top of ChatGPT and the API.
March 2026 · 8 releases
Gemma 4 (March 2026) is Google's current open-weight family — free, self-hostable, multimodal models from edge-sized (E2B) up to 31B, with 140+ language support.
GPT Image 2 (the DALL·E successor) lives inside ChatGPT — easiest to use conversationally, with strong instruction-following and in-image text.
GPT-5.4 mini is the cost-efficient workhorse — cheap, fast and good enough for the bulk of everyday traffic, the free/fallback tier in ChatGPT.
GPT-5.4 nano is the cheapest, fastest current-gen OpenAI model — an API-only micro-model for ultra-high-volume routing and classification, not deep reasoning.
Midjourney v8 is the aesthetics king — the most beautiful images out-of-box, but subscription-only with no free path for Indian users.
Mistral Small 4 (March 2026) is the value champion — a 119B/6.5B-active Apache-2.0 MoE that runs on one consumer GPU yet handles multimodal, reasoning and coding.
GPT-5.4 (March 2026) is the value-flagship: near-top quality on coding, research and math at roughly half the price of GPT-5.5, with native computer-use and a ~1M-token…
Gemini 3.1 Flash-Lite is Google's budget current-gen tier — rock-bottom pricing for high-volume, multimodal, latency-sensitive jobs.
February 2026 · 6 releases
1 Pro (Feb 2026) is Google's frontier model — top-tier reasoning, native multimodality (text/image/audio/video) and a 1M-token context, with India-friendly INR billing.
Claude Sonnet 4.6 (Feb 2026) is the everyday default — near-flagship coding at ~40% lower cost and ~2x the speed of Opus, with a 1M-token context and tunable thinking.
0 MoE that brings frontier reasoning and coding into self-hostable form.
Seedance 2.0 is ByteDance's leaderboard-topping video model — excellent realism and native audio, mainly reached via aggregator platforms.
Kling 3.0 is the value champion — near-flagship 4K quality and multilingual lip-sync at the lowest premium price, accessible from India.
ElevenLabs v3 is the expressive-voice leader — the most natural TTS and cloning, with a small free tier but commercial use behind paid INR plans.
January 2026 · 2 releases
GPT-5.3 Instant is the fast, non-reasoning default that free and Go-tier ChatGPT users actually talk to — great for everyday tasks, with no thinking delay.
Fish Audio S2 is the benchmark-topping open TTS — the most expressive self-hostable voice model, beating ElevenLabs on several independent tests.
December 2025 · 1 release
Runway Gen-4.5 is the filmmaker's toolkit — unmatched control and editing features, with predictable credit-based pricing.
November 2025 · 2 releases
FLUX.2 is the developer's photoreal image engine — API-first, with the strongest open-weight option for self-hosting.
Nano Banana Pro is Google's smartest native image model — superb conversational editing and text rendering, with a real free tier in India.
October 2025 · 2 releases
Claude Haiku 4.5 is the speed-and-cost tier — surprisingly capable coding (73% SWE-bench) at the lowest Claude price point.
Veo 3.1 is Google's safe, high-quality all-rounder — great visuals plus native audio, and actually accessible in India.
September 2025 · 3 releases
Sora 2 is OpenAI's cinematic video model — top-tier quality, but India-restricted and with its API being retired in late 2026.
Grok 4 Fast is xAI's budget / large-context tier — a 2M-token window at ~1/15th flagship price, the sensible default for most chat, RAG and agent workloads.
Magistral Small 1.2 (Sept 2025) is a 24B Apache-2.0 reasoning model with strong AIME math and transparent chain-of-thought that runs on a single consumer GPU.
July 2025 · 3 releases
Codestral 25.08 is Mistral's specialist code-completion / fill-in-the-middle model with a 256K context and tool use — fast, but proprietary-licensed for commercial use.
Qwen3-235B-A22B-Instruct is the mid-2025 open MoE instruct flagship (235B/22B, 256K) — a widely-deployed, self-hostable Apache-2.0 workhorse.
Qwen3-Coder-480B-A35B is Alibaba's flagship open coding MoE (480B/35B, 256K–1M context) — frontier agentic coding you can self-host under Apache 2.0.
May 2025 · 4 releases
DeepSeek R1-0528 is DeepSeek's last dedicated open reasoning model (671B/37B MoE) with top-tier AIME math and a family of practical small distills.
Gemini 2.5 Flash is a prior-gen value workhorse — very cheap multimodal reasoning with a 1M context, still hard to beat on raw price.
Gemini 2.5 Pro (May 2025) is Google's prior-generation flagship — still capable and stable with a proven 1M-token context, but outclassed by the Gemini 3.x line.
Imagen 4 is Google's reliability-and-text champion — cheap, accurate, with a genuinely usable free tier in India.
April 2025 · 5 releases
Qwen3-32B is the largest dense Qwen3 (32.8B, Apache 2.0) — a predictable, single-GPU-friendly model with hybrid thinking / non-thinking modes.
o3 is the last standing o-series reasoning model, now sunsetting — fine for legacy use but replaced by GPT-5.4/5.5 reasoning for anything new.
GPT-4.1 (April 2025) is a legacy 1M-context text model — cheap and capable, retired from the ChatGPT picker but still on the API for long-document text work.
Llama 4 Maverick (April 2025) is Meta's flagship open-weight model — a 400B-total / 17B-active MoE with multimodal input and GPT-4o-class quality you can self-host for…
Llama 4 Scout is the lightweight, single-GPU Llama 4 with a record 10M-token context — the most self-hostable open multimodal model for cost-constrained teams.
March 2025 · 1 release
Ideogram 3.0 is the typography specialist — when you need accurate, complex text inside an image (logos, posters, signage), this is the one.
January 2025 · 1 release
Sarvam is India's own speech stack — the most accurate Hindi / Indian-language STT + TTS, INR-billed and built for code-mixed Hinglish.
December 2024 · 1 release
Llama 3.3 70B Instruct is Meta's still-popular open-weight dense text model — GPT-4-class quality, 128K context, and the easiest large Llama to self-host.
October 2024 · 2 releases
Stable Diffusion 3.5 is the open-source workhorse — free, self-hostable and endlessly customisable, if not the absolute top on raw quality.
Whisper is the free open-source STT standard — multilingual and self-hostable, though Indian-language specialists can beat it on Hindi.
September 2024 · 1 release
Qwen2.5-72B-Instruct is the proven prior-gen dense 72B — widely fine-tuned and battle-tested, under the Qwen Licence rather than Apache 2.0.
May 2024 · 1 release
GPT-4o (May 2024) is the original multimodal workhorse, now legacy — kept mainly for vision/voice pipelines that newer cheap text models can't cover.
Want to know which of these suits your work? Try Find your AI, or line two up on Compare models.