Konkrete KI-Modelle und Modellfamilien im Überblick — wer sie entwickelt, wie sie sich in Reasoning, Coding, Multimodalität und Preis unterscheiden und wofür die einzelnen Linien im Alltag am besten taugen. Gegliedert nach Anbieter-Familien (GPT, Claude, Gemini …) und übergreifenden Einsatzzwecken wie Security, Bild & Video oder Open-Weight.
Anthropic released Claude Sonnet 5.5: $2 / $10 per million tokens, Opus-level benchmarks, but only at max effort. AA index 56 beats Astra.
OpenAI released GPT-6.1 Sol: $2 / $10 per million tokens, near-Astra coding performance, but behind Anthropic's models on the overall index.
Anthropic released Claude Opus 5.5: $4 / $20 per million tokens, about 40% cheaper to run than Opus 5 and ahead of Fable 5.1 on coding benchmarks.
Anthropic has ended the 50% boost. Since September 14 at 9:00 CEST, the permanent baseline is +25% – roughly 17% less capacity than before.
Zhipu AI ships GLM-5.3 — an open-weights MoE focused on coding and cybersecurity that leads agentic benchmarks among the open models.
xAI ships Grok 4.6 — a 500k-token context, image input and staggered pricing from $2/$6, built for long-running agentic work.
Meta ships Muse Spark 1.3 — successor to 1.2 with agentic coding and thought compression. What's new, what it costs, and who it's for.
DeepSeek ships V4.1-Flash: tops its own coding benchmarks, falls far behind on long-horizon agentic tasks — at a fraction of the competition's price.
Anthropic watermarks every model released from 2 August 2026. Fable 5.1 goes first. What it can do, where it fails, and who gets to verify.
On September 5 the weekly quota was back at 5% mid-session, plus a 50% boost until Sept 13. What to do with the capacity you just got handed.
OpenAI shipped GPT-6 Astra — the first model to hit the Critical tier of its Preparedness Framework. Almost nobody has tested it independently yet.
Google releases Gemini 3.8 Flash on September 2, 2026 — tuned for agents and long-horizon coding, same price, which doubles from January 2027.
Anthropic released Claude Fable 5.1 and Mythos 5.1. Base price holds, but cache reads are 75% cheaper and token usage drops by half.
July to August 2026 brought a wave of open models: GLM-5.3, DeepSeek V4-Pro, Qwen3.8. What the permissive licenses mean for self-hosting.
The 50 percent hike on Claude Sonnet 5 set for September 1 is off. Anthropic makes the 2/10 dollar introductory price permanent.
Since August 2026, Daybreak Blue gives vetted defenders access to Sol with fewer cyber refusals. It is not a new AI model.
Alibaba shipped Qwen3.8-Max on August 2, 2026: a 2.4T-parameter MoE, 1M-token context, image and video input via closed API.
Anthropic shipped Opus 5 — same token price as 4.8, and per the vendor near-Fable performance at roughly half the cost per task.
Moonshot AI launched Kimi K3 — an open-weight model with 2.8T parameters per the vendor. Open weights from July 27. What it means for the market.
OpenAI released GPT-5.6 (Sol/Terra/Luna) — but only after a pre-release review by the US government. What that means for frontier models.
Anthropic releases Claude Sonnet 5 on July 1, 2026 — near-Opus performance per the vendor, at an intro price of 2/10 dollars per million tokens.
Google Gemini is now live as an agent in boostN — work with your Google subscription through our execution pipeline. Mistral is coming next.
An export-control directive forced Anthropic to abruptly disable its top models Fable 5 and Mythos 5 for all users on June 12, 2026. Anthropic openly objects.
Anthropic discloses the numbers: Claude writes >80% of its production code. At the same time, the company argues for the option to pause frontier development.
Market overview mid-2026: Nano Banana 2, FLUX.2 and Midjourney V8 compared — which image tool fits which job for content and ad creatives.
Sora web/app off since 26 Apr 2026, API ends 24 Sep 2026. Kling, Runway and Vidu gain users. What it means for video workflows.
Anthropic ships Opus 4.8 — same price, cheaper fast mode, alignment near Mythos level. What another point release means for agencies working with it.
DeepSeek makes its 75% V4-Pro discount permanent: $0.435 input, $0.87 output per million tokens. What the price pressure means for model choice.
At Build 2026, Microsoft unveils seven in-house MAI models. MAI-Code-1-Flash runs right inside GitHub Copilot — cheaper than GPT-5.5.
MiniMax M3 launches as an open-weight model with sparse attention, a 1M context window and frontier-level coding benchmarks — at a fraction of the cost.
Alibaba ships Qwen 3.7 Max: 1M context, 35h autonomous, 1,158 tool calls. What the agentic flagship does and why vendor benchmarks call for caution.
Anthropic closed a $65B round at a $965B valuation, confidentially filed for an IPO, and is financing $36B of Google TPUs through debt.
Up to 75bn euros for 5 GW of AI capacity, Phase 1 at 45bn euros and 3.1 GW by 2031. Europe's AI infrastructure shifts toward Hauts-de-France.
Anthropic, OpenAI and Cursor are tilting flat rates toward metered billing. What has concretely changed and what hits you.
Per FT, Anthropic is negotiating a $50B round at $900B–$1T valuation. What the numbers say — and what they don't.
The Freiburg startup ships FLUX.2 [klein] under Apache 2.0 — sub-second inference on 13 GB VRAM and pressure on the big foundation-model players.
Cohere and Aleph Alpha merge into a $20B group. Schwarz Group commits $600M. What this means for sovereign AI in Europe.
GA since April 22, 2026: Google fuses Vertex AI, Agent Builder, and Memory Bank into the Gemini Enterprise Agent Platform — one stack for the lot.
Ahrefs shows just 38% of URLs cited in AI Overviews still rank in the top 10. The classic SEO playing field is shifting measurably.
675B total params, 41B active, Apache 2.0. Mistral Large 3 is the first true European frontier model with fully open weights.
OpenAI closed the round end of March at $852B valuation. Who got in, what Amazon, Nvidia and SoftBank pay, what it means.
OpenAI bundles Agent Builder, ChatKit, and Evals. Visual workflows replace boilerplate — ChatKit and Evals have been GA since DevDay 2025.
Since 7 May 2026: GPT-5.5-Cyber for vetted cybersecurity teams. Fewer refusals on vulnerability analysis, reverse engineering, detection engineering.
300 MW, 220,000 GPUs, doubled Claude Code limits. What the May 6, 2026 SpaceX deal means in practice for paying users.
At its devcon, Anthropic announced multi-agent fleets, outcomes-based control and a self-improvement loop for managed agents.
Ten templates for Wall Street workflows, Claude in Excel/PowerPoint/Word/Outlook, 600M company credit profiles directly inside Claude.
Since 5 May 2026, GPT-5.5 Instant is the new default model in ChatGPT. -52.5% hallucinations on law, medicine, finance — and fewer emojis.
Tiered discounts kick in automatically from $50,000 monthly spend. What changes for heavy users and when moving up the tier pays off.
Mistral ships Medium 3.5 at 77.6% SWE-Bench plus remote agents in Vibe and Le Chat. Local sessions can be teleported into the cloud.
Opus 4.6 dropped from the picker, Fast Mode now extra-usage only. Here's what shifted — and why almost nobody noticed.
Three independent changes degraded Claude Code from March through mid-April. Anthropic has disclosed the root causes and consequences.
OpenAI shipped GPT-5.5 and GPT-5.5 Pro on 23 April 2026. Pro lists at $30/$180 per MTok — six times the standard price. What that means.
Meta unveiled Muse Spark, the first model from its Superintelligence Labs — proprietary, not open. Little has been tested independently so far.
Mistral funds a dedicated AI data center near Paris. Target: 200 MW of European compute capacity by end of 2027 for sovereign AI.