The article reports the release of new AI models: Claude Opus 5.5 by Anthropic and GPT‑6 Sol and GPT‑6 Luna by OpenAI, noting that GPT‑6 variants are priced at half the cost of their GPT‑5.6 counterparts. It provides a detailed pricing table comparing input, cached input, and output costs across several models, highlighting how GPT‑6 Luna is among the cheapest ever offered by OpenAI. The author also comments on visual differences in model outputs, noting that GPT‑6 outputs are more muted compared to GPT‑5.6.
Mistral has released a preview of its new Mistral Large 4 model, a 1 trillion‑parameter, 49 billion‑active‑parameter language model trained on a cluster of 3,800 NVIDIA Grace‑Blackwell GPUs. The preview is available through their API, with two reasoning levels—"none" and "high"—and the company plans to release the open‑weights version by the end of the month. In preliminary tests, the model scores 38 on Artificial Analysis, outperforming last year’s Mistral Large 3 and approaching the performance of larger competitors.
Qwen3.8-Flash-Next is an open‑weights multimodal Mixture‑of‑Experts (MoE) model previewing the architecture of Qwen4. It contains 125 B tokens with only 6 B active, giving a performance boost. The author has tested it on a DGX Spark with Unsloth quantized models, exploring variants like UD‑IQ1_S and UD‑Q2_K_XL, and highlighted a high‑reasoning‑effort example from UD‑Q2_K_XL.
DeepSeek V4 Pro 0813 (on OpenRouter) The latest DeepSeek Pro model is now available, via API only. I had to link to OpenRouter because DeepSeek don't have any obvious announcement page for their new model.
Friday's big release was Qwen 3. 8 27B , an Apache 2 licensed 27B parameter vision-capable LLM from Alibaba's Qwen research lab.
Simon Willison announces the release of ttok 1.0 after discovering that the previous version defaulted to the GPT‑4 tokenizer instead of the expected GPT‑5/GPT‑6 tokenizer. He explains that switching the default was a suitable reason to ship the new version. The release notes reference an experiment by William Liu that confirms the tokenizers for GPT‑5 and GPT‑6 models are likely identical, as all seven models report the same token count across 31 test fixtures.
Qwen 3. 8 27B scores 52 on the Artificial Analysis Intelligence Index That's the same score as GPT-5.
Claude Haiku 5.5 is Anthropic’s new fast, low‑cost model that matches GPT‑6 Luna’s pricing of $0.10 per 1,000 input tokens and $0.50 per 1,000 output tokens up to 100,000 tokens, with a 5× increase beyond that threshold. It uses a less generous tokenizer, meaning the same long prompt consumes about 1.25× more tokens than Haiku 4.5, which raises hidden costs. Benchmarks show Haiku 5.5 outperforms Luna within the 100,000‑token limit, but Luna becomes more economical for larger workloads.
GPT‑6 Astra is a new OpenAI model rolling out today to a limited set of organizations and soon to all ChatGPT Plus, Pro, Business, Enterprise users, and via the OpenAI API and AWS. It is priced at $10/million input and $50/million output, matching Claude Fable 5/5.1, and outperforms Fable on most OpenAI self‑reported benchmarks, achieving 99.9% on the ARC‑AGI 3 benchmark with a custom Provider Adapter harness. Astra excels in security tasks—scoring 100% on ExploitBench, 42.4% on ExploitGym, and 99.2% on SRE‑Bench—and handles long context well, hitting 100% on OpenAI’s eight‑needle benchmark at 256K–512K tokens and 96.3% at 512K–1M tokens, though it remains behind Fable on the Intelligence Index and Meta’s Muse Spark 1.3.
The release of llm-gemini 0.34 introduces the new Gemini 3.8‑Flash model, available in low, medium, and high thinking levels, and fixes an issue where async responses failed to record the resolved model version. The update also notes that Google has released Gemini 3.8‑Flash (and a restricted 3.8 Flash Cyber version) today, with example outputs (pelicans) demonstrating the model’s performance across the different thinking levels. The author highlights Gemini Flash’s speed, low cost, and competence in generating HTML, JavaScript, and Markdown‑SVG content, citing a 13‑second, 1.8‑cent example of an HTML output.
The article reports that on a set of 100 randomly selected tasks from an internal Binary Exploitation benchmark, GLM‑5.3 achieved full control‑flow hijacks in 4% of the trials, while Claude Mythos Preview did so in 6%. Both models outperform earlier versions such as Claude Opus 4.6 and GLM‑5.2, which succeeded in none of the trials. This indicates that a significant threshold in adversarial exploitation capabilities has been crossed by the newer models.
The article announces the release of llm-meta-ai 0.2, featuring Meta AI’s new Muse Wonder and Muse Wonder Full models, described as the most capable models yet. These models are suggested to be Meta’s first Fable‑class models, with Artificial Analysis scoring Muse Wonder (Max) equivalent to GPT‑6.1 Sol. The author updated the llm-meta-ai plugin to support the new model and provides pelican‑drawing benchmarks across six reasoning levels, noting significant improvements over previous releases.