Claude’s maker is building an in-house custom-silicon team. This suggests the next AI advantage may come from coordinating models, software, and hardware as one system.
Demis Hassabis moves to chair and Alphabet chief scientist. Koray Kavukcuoglu takes day to day control of Gemini. Jeff Dean is leaving after 27 years. Read together, it is an admission that research and shipping run on different clocks.
3 min read →
Llama 5 was never released, llama.com now redirects, and Meta Superintelligence Labs shipped a closed model in its place. The open weight frontier is now DeepSeek, Qwen, Kimi, and Mistral.
3 min read →
OpenAI cut GPT-5.6 Luna by 80 percent on 30 July. Grok 4.5 claims to use a quarter of the tokens of its rivals. DeepSeek charges a fiftieth of the going rate on cached input. Capability has stopped being the only axis of competition.
3 min read →
An editorial snapshot, not a measured global ranking.
Reviewed and reordered August 7, 2026.
How we rank →
An editorially selected map of what moved across the AI landscape, one signal per beat.
Free frontier access for researchers is cheap distribution for OpenAI and genuinely useful for labs without budget. Both…
1 min read → 02 ModelsOpen is doing a lot of work in AI marketing. The licence, not the download button, decides what…
3 min read → 03 InfrastructureIf the next advantage in AI comes from tuning models and silicon together, then a lab that only…
2 min read → 04 EnterprisePhysical AI carries a higher cost of error than a chatbot. Releasing a reasoning model for the rare…
1 min read → 05 ApplicationsAlmost everyone assumes their AI subscription is private by default. On most products it is the opposite, and…
4 min read → 06 MarketsWhen every flagship has a million token context and comparable benchmark scores, the differentiator becomes what a finished…
3 min read → 07 PolicyA 100 million dollar programme across up to ten hubs will not close the compute gap, but it…
1 min read → 08 TalentGoogle has the research depth and a Pro tier that has not moved since February. This reorganisation is…
3 min read →Reporting says what happened. These are the open questions our reporting raised, each with a specific answer coming and a way to tell when it arrives. Not predictions.
What would settle it Anthropic lists USD 2.00 input and USD 10.00 output per million as an introductory rate ending 31 August 2026, after which it becomes USD 3.00 and USD 15.00. The pricing page on 1 September settles it.
What would settle it Alibaba announced the model on 3 August 2026 with a commitment to publish weights within a week. Weights appearing under a named licence would settle it. Until then we do not describe the model as open.
What would settle it It launched on 19 February 2026 and still ships under a preview model identifier. A stable identifier from Google settles it, and it is the clearest single test of whether the August leadership split changes the shipping cadence.
Side by side intelligence that replaces feature overload with a confident choice.
Drafting, editing, synthesis, and evidence-led research with citations you can check.
13 reviewed choices →Code generation, debugging, refactoring, and agents that work across a whole repository.
14 reviewed choices →Images, brand and design systems, presentations, and short-form video.
7 reviewed choices →Spreadsheets, long documents, statistics, and decisions that need a defensible answer.
11 reviewed choices →Strategy, operations, sales, support, and the everyday admin that eats a week.
5 reviewed choices →Agents, integrations, and repeatable multi-step workflows that run without you.
12 reviewed choices →Ask one question and several AI systems answer independently, review each other, and hand you a synthesis that shows where they agree and where they do not. You choose the direction. One Conductor is in development at 18BYTE Product Studio.
ONE Collective | ONE Direction | ONE Outcome.
One Conductor is a product of 18BYTE Product Studio, which also publishes 18BYTE. Coverage that involves it is clearly disclosed.
