Release archive
The timeline.
Every model we track—released, announced or still a signal on the horizon.
Anthropic · FEB 05, 2026
Reasoning-focused flagship aimed at long-running agent workflows.
$15.00/Mtok
No reviewed reports yet.
Read more →Z.AI · FEB 15, 2026
Open-weights flagship from Z.AI, strong on tool use for its size.
$0.60/Mtok
No reviewed reports yet.
Read more →MiniMax · FEB 15, 2026
Long-context model with an emphasis on cost per token.
No reviewed reports yet.
Read more →Alibaba · FEB 16, 2026
Sparse mixture-of-experts release in the Qwen3.5 line.
No reviewed reports yet.
Read more →Google DeepMind · FEB 19, 2026
Native multimodal model with a very large context window.
$7.00/Mtok
No reviewed reports yet.
Read more →OpenAI · FEB 24, 2026
Code-specialised variant tuned for repository-scale edits.
No reviewed reports yet.
Read more →OpenAI · MAR 06, 2026
General-purpose update with improved tool-use reliability.
$10.00/Mtok
No reviewed reports yet.
Read more →Z.AI · MAR 15, 2026
Z.AI describes GLM-5-Turbo as a proprietary, latency-optimized variant of GLM-5 built specifically for agent workflows, with stronger tool use and stability across long execution chains than the open-weight base model.
$1.20/Mtok
No reviewed reports yet.
Read more →OpenAI · MAR 17, 2026
Smallest GPT-5.4 tier, aimed at high-volume classification.
No reviewed reports yet.
Read more →Xiaomi · MAR 18, 2026
Xiaomi's multimodal model, positioned for on-device work.
No reviewed reports yet.
Read more →Z.AI · APR 07, 2026
Point update to GLM-5 with a longer context window.
No reviewed reports yet.
Read more →ByteDance · APR 14, 2026
ByteDance describes Seedance 2.0 as a video model built around combined audio-and-video generation from mixed text, image, audio, and video inputs, with director-style control over camera work and lighting.
No reviewed reports yet.
Read more →Anthropic · APR 16, 2026
Improvement on Opus 4.6 in advanced software engineering, with the largest gains on the hardest tasks.
$5.00/Mtok
No reviewed reports yet.
Read more →Moonshot AI · APR 20, 2026
Agentic-focused release in the Kimi K2 line.
No reviewed reports yet.
Read more →Xiaomi · APR 22, 2026
Open-weights 1T-parameter mixture-of-experts model, pitched at frontier-tier scores for a fraction of the cost.
No reviewed reports yet.
Read more →DeepSeek · APR 24, 2026
Open-weights reasoning model at an aggressive price point.
$0.90/Mtok
No reviewed reports yet.
Read more →DeepSeek · APR 24, 2026
Smaller, faster sibling to V4 Pro — 284B total / 13B active, 1M context, thinking and non-thinking modes.
No reviewed reports yet.
Read more →OpenAI · APR 24, 2026
Flagship update with a substantially larger context window.
No reviewed reports yet.
Read more →Alibaba · MAY 17, 2026
Agent-focused flagship of the Qwen3.7 line, served through Alibaba Cloud Model Studio.
No reviewed reports yet.
Read more →Google DeepMind · MAY 19, 2026
First of the Gemini 3.5 family — flagship-level intelligence at Flash speed, built for agents.
No reviewed reports yet.
Read more →Anthropic · MAY 28, 2026
Further reasoning gains over Opus 4.6 on long-horizon tasks.
No reviewed reports yet.
Read more →Anthropic · JUN 09, 2026
A Mythos-class model made safe for general use; state-of-the-art on nearly all tested capability benchmarks. Briefly suspended over export controls, redeployed 1 July.
$10.00/Mtok
- CodingClaude Fable 5.1 High vs Fable 5 high in claude code
Moonshot AI · JUN 12, 2026
Open-weights coding model — around 30% fewer reasoning tokens than K2.6 at higher scores on Moonshot's own coding bench.
No reviewed reports yet.
Read more →Sakana AI · JUN 23, 2026
Sakana AI's frontier model, developed in Tokyo.
No reviewed reports yet.
Read more →Anthropic · JUN 30, 2026
Mid-tier sibling to Opus, tuned for latency-sensitive production work.
$3.00/Mtok
No reviewed reports yet.
Read more →OpenAI · JUL 09, 2026
Agentic tier of the GPT-5.6 family.
- CodingUnder review: claimed to fix a race condition in a submitted snippet, reporter says the fix looks plausible but is still testing it under load.
xAI · JUL 16, 2026
xAI's smartest model for coding, agentic tasks and knowledge work; trained alongside Cursor.
$2.00/Mtok
No reviewed reports yet.
Read more →Moonshot AI · JUL 16, 2026
New generation of Moonshot's open-weights line.
$0.50/Mtok
No reviewed reports yet.
Read more →Google DeepMind · JUL 21, 2026
Cybersecurity-tuned model deployed through the CodeMender agent in a limited government and partner pilot.
No reviewed reports yet.
Read more →Google DeepMind · JUL 21, 2026
Fastest, most cost-effective 3.5-class model, serving around 350 output tokens per second.
$0.30/Mtok
No reviewed reports yet.
Read more →Google DeepMind · JUL 21, 2026
Workhorse Flash model with better coding, knowledge work and multimodal performance.
$1.50/Mtok
No reviewed reports yet.
Read more →Anthropic · JUL 24, 2026
Anthropic's flagship model, positioned for long-horizon agentic work and coding.
SWE-bench Verified 96.0%SWE-bench Pro 79.2%$15.00/Mtok
- CodingVideo game built with opus 5
Alibaba · AUG 02, 2026
Largest tier of the Qwen3.8 series.
- OtherSubmitted report awaiting review — task category and takeaway not yet verified against the source.
Meta · AUG 06, 2026
Meta Superintelligence Labs' flagship LLM and Llama successor; powers the Meta AI assistant.
No reviewed reports yet.
Read more →xAI · AUG 12, 2026
Builds on Grok 4.5 with a focus on long-running agents and more ambitious interactive and visual work.
$2.00/Mtok
No reviewed reports yet.
Read more →Google DeepMind · AUG 13, 2026
Cost-optimised Gemini tier with the full context window.
$0.40/Mtok
No reviewed reports yet.
Read more →Z.AI · AUG 14, 2026
Third point release in the GLM-5 series.
- OtherGLM 5.3 flash with blender to generate video
DeepSeek · AUG 21, 2026
Experimental multimodal build of V4 Flash, adding vision input through the API.
No reviewed reports yet.
Read more →Z.AI · AUG 26, 2026
First natively multimodal GLM-5 release — 320B total / 18B active, MIT-licensed, tested publicly as “Ox Alpha”.
- Other@TheAhmadOsman: Just a reminder that GLM 5.3 Flash, DeepSeek V4.1 Flash, Qwen 3.8 Next Flash, and even Qwen 3.8 27B are all outperforming (in both intelligence and capabilities) every model that was considered "frontier intelligence" in Xmas 2025 (just 10 months ago) Opensource AI is on fire
Alibaba · AUG 26, 2026
Open-weights 125B mixture-of-experts model (6B active), an early preview of the Qwen4 architecture.
No reviewed reports yet.
Read more →Anthropic · SEP 01, 2026
Newest Mythos-class model, with gains in cybersecurity and biology. The same underlying model as Fable 5.1 under a different safeguard layer; access remains limited to a small set of vetted organizations through Project Glasswing.
Terminal-Bench 4.0 60.9%$10.00/Mtok
No reviewed reports yet.
Read more →Anthropic · SEP 01, 2026
Anthropic's most advanced model for coding and knowledge work, with the largest gains on terminal-based scientific and engineering work, computer use, and long-horizon agentic tasks. Around 25% cheaper than Fable 5 for typical workloads, with cache reads down 75%.
Terminal-Bench 4.0 55.8%Terminal-Bench-Science 0.1 52.6%$10.00/Mtok
- Other@argofowl: astra has been pretty solid for me after i took my time with the agents.md it's not perfect, nothing is, but i like it a bit more than fable 5.1 right now controversial i know
- Codinglaude Fable 5.1 has gone absolutely unhinged with the visual aesthetic
- CodingClaude Fable 5.1 ONE SHOTTED this Call of Duty game.
Google DeepMind · SEP 02, 2026
Google's workhorse Gemini 3.8 model for coding, agentic tasks and multi-step reasoning, launched at the introductory price of $0.75 input and $3.75 output per million tokens.
$3.75/Mtok
No reviewed reports yet.
Read more →Meta · SEP 02, 2026
Meta's updated Muse Spark model, with improved coding, agentic workflows and native multimodal perception.
No reviewed reports yet.
Read more →Google DeepMind · SEP 02, 2026
Cybersecurity-focused Gemini 3.8 variant for vulnerability discovery and automated patching, available to trusted defenders through Google's Fairwind Program.
No reviewed reports yet.
Read more →OpenAI · SEP 03, 2026
First model in the GPT-6 generation, which OpenAI describes as a new frontier on computer and browser use and its best model for software engineering to date, citing bug detection, terminal tasks and codebase analysis. It is also the first model OpenAI has designated as reaching the Critical cybersecurity threshold under its Preparedness Framework, and it ships under added safeguards, with the widest access limited at launch to vetted testers in the Daybreak program.
$10.00/Mtok
- CodingOpus 5.5 vs. GPT-6 Astra, cost comparison. We paired both of these with Tesseract to recreate our launch video.
- CodingI asked GPT-6 Astra to make a portfolio inside a console similar to a Gameboy and LOOK AT THIS
- CodingI vibe coded for 8 hours straight on live stream yesterday. GPT 6 Astra is unreal.
OpenAI · SEP 08, 2026
The faster GPT-Image-2.5 API tier, with improvements to image quality, editing and generation speed.
No reviewed reports yet.
Read more →OpenAI · SEP 08, 2026
The higher-precision GPT-Image-2.5 API tier for detailed creative work, trading generation time for control and fidelity.
No reviewed reports yet.
Read more →OpenAI · SEP 10, 2026
Full-duplex voice model for the API that handles simultaneous listening and speaking while delegating deeper reasoning and tool calls to a backend model.
No reviewed reports yet.
Read more →DeepSeek · SEP 10, 2026
DeepSeek's faster V4.1 architecture, replacing V4 Pro traffic in the hosted API while adding native multimodal support.
- Other@TheAhmadOsman: Just a reminder that GLM 5.3 Flash, DeepSeek V4.1 Flash, Qwen 3.8 Next Flash, and even Qwen 3.8 27B are all outperforming (in both intelligence and capabilities) every model that was considered "frontier intelligence" in Xmas 2025 (just 10 months ago) Opensource AI is on fire
Sakana AI · SEP 11, 2026
Sakana's cost-oriented model-orchestration system, dynamically routing tasks across a larger pool of open and specialized models.
No reviewed reports yet.
Read more →Sakana AI · SEP 11, 2026
Second-generation Fugu Ultra orchestration system, aimed at raising peak capability without depending on a single frontier model.
No reviewed reports yet.
Read more →Moonshot AI · SEP 11, 2026
Kimi Code preview with more efficient reasoning, adjustable thinking effort and a 1M-token context window, served behind the existing kimi-for-coding model ID.
No reviewed reports yet.
Read more →Google DeepMind · SEP 15, 2026
Google's cost-efficient real-time dialogue model, combining fluid voice interaction, visual grounding and asynchronous function calling.
No reviewed reports yet.
Read more →Google DeepMind · SEP 15, 2026
Higher-reasoning Gemini Live tier that can work through multi-step tasks while maintaining an uninterrupted spoken conversation.
No reviewed reports yet.
Read more →Alibaba · SEP 18, 2026
Native omni-modal model accepting text, image, audio and video with up to 1M tokens of context, built on the Qwen3.8-Flash-Next architecture.
No reviewed reports yet.
Read more →Xiaomi · SEP 21, 2026
The efficiency-balanced MiMo V2.6 checkpoint, a 311B open-weight omni-modal model trained across coding, agentic, visual and cybersecurity tasks.
No reviewed reports yet.
Read more →xAI · SEP 21, 2026
xAI's frontier model for coding, agentic tasks and knowledge work, with stronger long-running task performance and self-verification.
$6.00/Mtok
- Other@bridgebench: We gave 4 models the same prompt: build a 3D rocket launch. GPT 6 Luna: <$0.01, 1 minute GPT 6 Sol: $0.11, 1 minute Grok 4.7: $0.29, 11 minutes Claude Opus 5.5: $1.52, 12 minutes GPT 6 Luna built a rocket launch scene in 65 seconds for less than a penny. The cost of frontier AI is collapsing in real time.
Xiaomi · SEP 21, 2026
The larger MiMo V2.6 checkpoint, combining text, image, video and audio input with a 1M-token context window for long-running coding and agent work.
No reviewed reports yet.
Read more →OpenAI · SEP 22, 2026
OpenAI's efficient reasoning model for focused, high-volume workloads, with text and image input through the Responses and Chat Completions APIs.
$0.50/Mtok
- Other@bridgebench: We gave 4 models the same prompt: build a 3D rocket launch. GPT 6 Luna: <$0.01, 1 minute GPT 6 Sol: $0.11, 1 minute Grok 4.7: $0.29, 11 minutes Claude Opus 5.5: $1.52, 12 minutes GPT 6 Luna built a rocket launch scene in 65 seconds for less than a penny. The cost of frontier AI is collapsing in real time.
Anthropic · SEP 22, 2026
Anthropic's leading model for agentic coding, computer use and knowledge work, with lower costs and faster generation than Opus 5.
$20.00/Mtok
- Other@_MaxBlade: Opus 5.5 built a multiplayer lawn mow simulator with beers and cigars Made it live on my server and chat joined. the future is looking insane.
- Vision@ishuagra02: Oh my god... Opus 5.5 is the best model in 3D rendering, BY FAR. I gave the following prompt: "create a highly detailed 3D scenery that faithfully captures this photo in a cel-shaded cartoon art style" and it crushed it. The detail, the lighting, the colors, the visual fidelity... I have no words for this. Didn't even exceed the Claude Pro session limits. Cost around $38 in API usage.
- Other@bridgebench: We gave 4 models the same prompt: build a 3D rocket launch. GPT 6 Luna: <$0.01, 1 minute GPT 6 Sol: $0.11, 1 minute Grok 4.7: $0.29, 11 minutes Claude Opus 5.5: $1.52, 12 minutes GPT 6 Luna built a rocket launch scene in 65 seconds for less than a penny. The cost of frontier AI is collapsing in real time.
OpenAI · SEP 22, 2026
OpenAI's reasoning model for complex coding and agentic workflows, with text and image input through the Responses and Chat Completions APIs.
$10.00/Mtok
- Other@bridgebench: We gave 4 models the same prompt: build a 3D rocket launch. GPT 6 Luna: <$0.01, 1 minute GPT 6 Sol: $0.11, 1 minute Grok 4.7: $0.29, 11 minutes Claude Opus 5.5: $1.52, 12 minutes GPT 6 Luna built a rocket launch scene in 65 seconds for less than a penny. The cost of frontier AI is collapsing in real time.
Fireworks AI · SEP 23, 2026
A specialized reasoning model based on Kimi K3 that Fireworks says preserves comparable quality while using roughly 40% fewer output tokens.
$15.00/Mtok
No reviewed reports yet.
Read more →Perceptron · SEP 25, 2026
Perceptron's public model for embodied agents, adding native audio, video tracking, web search, sub-agent calls and more complex visual reasoning.
No reviewed reports yet.
Read more →Anthropic · SEP 28, 2026
A faster, lower-cost complement to Opus 5.5 for everyday tasks, bug fixes and polished documents, slides and spreadsheets.
Terminal-Bench 4.0 70.6%FrontierCode 1.1 (Max) 46.2%$10.00/Mtok
- Agentic@_MaxBlade: sonnet 5.5 just built the most addicting game I have ever played... 🤯 only up clone. fully multiplayer and running in the browser, i made it over 200m before falling. link below, how far can you go?
- Other@filicroval: what the fuck, why is Sonnet 5.5 this good?? gave Sonnet and Opus 5.5 the exact same prompt. Sonnet came back with this masterpiece. all in a SINGLE HTML file. Sonnet won this one, it's not even close
- Agentic@_MaxBlade: I asked Sonnet 5.5 to create a overcooked clone i could play in the broswer. the results shocked me... It is delivering triple aaa level polish at 20x cheaper than opus 5.5 if you put it in x-high it will RUN. this model will be a perfect worker agent for opus 5.5 or the upcoming fable 5.5 to orchestrate. Anthropic is MOGGGING rn.
OpenAI · SEP 29, 2026
A major upgrade to GPT-6 Sol that OpenAI says brings near-Astra performance to agentic coding, computer use and professional work at one-fifth of Astra's standard token prices.
$10.00/Mtok
No reviewed reports yet.
Read more →Alibaba · OCT 02, 2026
A unified text-to-image generation and editing model with native transparent-image support, multi-reference editing and local edit controls.
No reviewed reports yet.
Read more →Google DeepMind · OCT 06, 2026
Google's high-efficiency image generation and conversational editing model, with stronger prompt adherence, character consistency, text rendering and panoramic output.
No reviewed reports yet.
Read more →Mistral AI · OCT 06, 2026
Mistral's largest model to date, a natively multimodal mixture-of-experts model for coding, agentic workflows and enterprise work with a 1M-token context window.
$4.18/Mtok
No reviewed reports yet.
Read more →Anthropic · OCT 07, 2026
Anthropic's fastest small model, designed for high-volume, cost-sensitive work such as summaries, compaction, classification, subagents and latency-sensitive browser use.
OSWorld 2.1 (offline subset) 72.4%Humanity's Last Exam (tools) 57.4%$0.50/Mtok
No reviewed reports yet.
Read more →Alibaba · OCT 09, 2026
The cost- and speed-oriented Qwen-Image-2.1 tier, with unified generation and editing, native transparency, multiple references and local controls.
No reviewed reports yet.
Read more →Google DeepMind · OCT 13, 2026
Google's frontier reasoning model for long-horizon work across coding, finance, legal, cybersecurity and science, with a unified 1M-token input-output limit.
$10.00/Mtok
No reviewed reports yet.
Read more →