GPT-6 Sol and Luna landed on September 22, 2026, and the interesting number is not on any benchmark chart. It is on the pricing page. GPT-6 Sol costs $2.00 per million input tokens and $10.00 output. The model it replaces, GPT-5.6 Sol, is $4.00 and $20.00. That is half, on both sides of the meter. The capability story is thinner. The only independent measurement published on launch day puts GPT-6 Sol one point above its predecessor, and GPT-6 Luna two points below its own.

The Vercel AI SDK is a free, open-source TypeScript toolkit for building applications that talk to large language models. It gives you one consistent set of functions for calling a model, streaming its output, wiring up a chat interface, and letting the model run your own code. This walkthrough covers what the toolkit includes, then builds a small LLM-powered app step by step: project setup, a first model call, a streamed response, and a tool call the model can trigger on its own.

On September 3, 2026, ARC Prize published two scores for the same model on the same benchmark on the same day. GPT-6 Astra scored 62.7% on arc-agi-3 under the standard, provider-neutral harness, and 99.9% under a harness supplied by OpenAI. Both numbers are real. Both were produced by the organization that runs the benchmark. ARC Prize says it will report both, labeled.

Muse Code is Meta's terminal coding agent, released in beta on 5 August 2026 by Meta Superintelligence Labs. It runs on Muse Spark 1.2, a coding-focused model that shipped the same day, and Meta is selling it on price rather than on capability. That is not a critic's reading. It is Mark Zuckerberg's own framing, and it is the most interesting thing about the release, because it tells you where coding agents are in their product cycle.

DeepSeek V4 Flash got an update on July 31, 2026, and the number doing the work in the coverage is 82.7 on Terminal-Bench 2.1, which would place it among the best coding models in the world. That number is self-reported, its run conditions are partly undisclosed, and no DeepSeek model appears anywhere on the benchmark's own leaderboard. The genuinely interesting thing in the release is in the same changelog entry, one sentence further down, and it has nothing to do with benchmarks.

GPT-6 Astra landed on September 3, 2026, and the announcement page is doing several jobs at once. Some of what it describes is generally available in the API today. Some is capability-complete but gated behind a limited rollout. Some is behind an opt-in flag in a config file. And some is a name attached to a plan with no date, no criteria, and no definition.

Anthropic released Claude Fable 5.1 on September 1, 2026 saying it "will cost an estimated 25% less than Fable 5 for typical workloads," with savings up to roughly 45% on highly agentic work. In the week since, three independent measurements have been published, and they do not agree. One found the model costs about 20% more per task than its predecessor.

Muse Glimmer is a 30-billion-parameter open-weight model Meta released on August 10, 2026, under the Apache 2.0 license. It is built for agentic work, it accepts images as well as text, and a quantized build fits in 24 GB of video memory, which means it runs on a single consumer graphics card or a well-specified Mac. The license is the part that deserves attention, and the benchmark table Meta published alongside it is more interesting than the announcement.

xAI released Grok 4.6 on August 12, 2026. Within a week it was available in GitHub Copilot and on Amazon Bedrock, and the company pushed its Build Mode agent surface to general availability. The model itself is an incremental step over Grok 4.5, which shipped four weeks earlier. The release is worth attention, but not primarily for the model. The price and the distribution are the story, and the benchmark table xAI published is more mixed than the announcement's framing suggests.

Gemini 3.7 Flash shipped on August 13, 2026, three weeks after 3.6 Flash. The coverage has focused on the benchmark jump, which is real. It has almost entirely missed the pricing footnote, which is the part that will actually appear on somebody's invoice. The number everyone is quoting is a promotion with an expiry date, and the expiry is four and a half months away.

The reported Stripe OpenRouter acquisition is being covered as a valuation story. It is not. The number is the least interesting thing about it, and the part worth your attention is what a payments company gets to see once it owns the wire between applications and models. Start with what is established and what is not, because the two are being blurred.

Between July 21 and August 5, 2026, four separate labs published a coding benchmark figure that the technology press reprinted as though it were a position on a leaderboard. None of those four numbers appears on the leaderboard in question. This is not a story about dishonesty, and the accusation framing gets it wrong. It is a story about a measurement pipeline with no verification step at the vendor end, and…

Qwen3.8-Max reached general availability on 3 August 2026, two weeks after the preview we covered in our look at Qwen3.8. That post ended with a checklist: five specific things that needed to land before the model could be treated as a frontier-class option rather than a promising preview. Three arrived. Two did not, and they are the two that matter most.

Laguna S 2.1 is poolside's open-weight coding model, released on 21 July 2026 with weights on Hugging Face under the OpenMDW-1.1 license. It is a mixture-of-experts model at 118 billion total parameters with 8 billion active per token, and a context window of just over a million tokens. The number that earns it attention is 70.2 percent on Terminal-Bench 2.1, which poolside reports beats open models roughly ten times its size.

Open vs closed AI models is the choice that now sits under most serious AI decisions: do you run a model you can download and control, or rent one through a vendor's API? The honest framing has three tiers, not two. A model can be fully open source (weights, training data, and training recipe all published), open weight (the trained weights released, but the data and recipe held back), or closed and proprietary (available only as a hosted service).