Models & Releases
Deep, evidence-first coverage of major model releases and the labs behind them: capabilities, pricing, benchmarks, and what each release changes for builders.
Models & ReleasesA 7B Model Beat Claude Opus by Routing, Not Reasoning
Routing across a pool of frontier models is a credible architecture bet in 2026, but its production wins are still unproven. Here's how to interrogate them.
Models & ReleasesClaude Fable 5 Returns With a Government Key in the Lock
Eighteen days of outage turned a coding model into an eschatology. Here's what the saga actually revealed about who controls frontier AI.
Models & ReleasesThe "US" vs Them: Fable Off, GPT-5.6 Gated
Washington flipped off Fable for the planet, then opened GPT-5.6 to twenty vetted U.S. orgs. Frontier AI access is now a sovereignty variable.
Models & ReleasesGPT-5.6 Deployment Starts Behind a Federal Gate
Sol, Terra, and Luna rewrite OpenAI's stack economics, but the first wave is customer-by-customer vetting, not a normal API launch.
Models & ReleasesFable 5 Went Dark. Praxis Is the City Built to Outrun It.
A single Commerce Department letter switched off Anthropic's Fable 5 worldwide; Praxis wants to make sure the next frontier model can't be turned off the same way.
Models & ReleasesFable 5 Biology Restrictions Have a Real Job
The fallback to Opus 4.8 is best understood as a frontier-access control system, with real consequences for biotech teams and AI drug discovery workflows.
Models & Releases14 Days of Fable 5: The Shutdown That Rewired AI
How a defensive cybersecurity preview became the most powerful public AI model, triggered an export-control emergency, vanished worldwide, and returned under restrictions.
Models & ReleasesFable Without Fable: Sakana Fugu Ultra's Big Bet
Sakana's most interesting move is selling learned orchestration as a frontier-model substitute, with Fable-class pressure and very different production risks.
Models & ReleasesThe Fable 5 Mythos 5 Export Directive Hit Your API
The U.S. Did more than pause a model; it turned model access into an availability risk engineers have to design around.
Models & ReleasesLong-Horizon Agents Run for Hours. Wield Them Safely
Fable 5 migrated 50 million lines of Stripe code in a day. The skill that matters now is objective delegation plus containment, not prompt engineering.
Models & ReleasesGPT-5.4 Drug Discovery: AI Improves a Lab Reaction
Paired with Molecule.one's Maria AI and an automated lab, GPT-5.4 picked the problem, proposed a counterintuitive additive, and 10,080 reactions later it held up.
Models & ReleasesFable 5 Export Controls: A New Model-Recall Precedent
The first US export-control recall of a live frontier model just rewrote your model-dependency risk model.
Models & ReleasesThe Magic They Switched Off: Get Your Claude Max Ready for Fable 5
For 72 hours we held the most powerful model ever shipped. Then Washington switched it off. This is how to build a Claude Max setup ready to wield Fable the hour it returns.
Models & ReleasesUS Blocks Foreign Access to Anthropic's Fable 5 and Mythos 5
A Commerce Department export-control directive bars all foreign nationals from Anthropic's two most advanced models — and forced the company to switch them off for everyone.
Models & ReleasesOpen-Source Reasoning Models in 2026: The Gap Has Closed
DeepSeek-R1, Qwen 3, and Llama 4 put frontier-class reasoning within single digits of proprietary models, at 10 to 30 times lower cost.
Models & ReleasesAnthropic S-1 IPO: What's Confirmed vs. The $965B Leak
The confirmed ledger on Anthropic's IPO is one sentence long. Everything else, including the $965 billion valuation, is anonymous-source reconstruction that history says gets revised.
Models & ReleasesBest Local LLM for Coding on 16GB VRAM: June 2026 Rankings
We ran the quantized contenders ourselves: Gemma 4 12B and JetBrains Mellum 2 lead the 16GB tier, and the gap to hosted Claude is exactly quantifiable.
Models & ReleasesReasoning-First LLMs: Make Models Reason, Not Rationalize
Your model's chain of thought is a narrative, not a derivation. Here is the stack that forces it to actually compute the answer.