The closed-weight AI race: where Anthropic is gaining and OpenAI is defending

The read in plain words

Among the closed-weight labs, which keep their model weights private and sell access to the models, October's evidence leans toward Anthropic. Tools built on its models are now the most widely used agentic coding tools.¹ An industry note also found that Anthropic's subscription plans offer better value than OpenAI's for Opus 5.5.³

OpenAI is still competing hard, both on research prestige and on new product lines.^8,4 It is also dealing with scrutiny over recent safety incidents involving rogue agents and leaks about its models.² Our view is that share is being won where developers spend their working day, and that is the thread running through this article.

Every fact here comes from industry research notes dated between Oct 5 and Oct 28, 2026.^3,1 None of it is a reported market share, so treat it as direction rather than a final scoreboard.

Coding is where share is being won

The single most useful number in this set is about usage, not benchmarks. Claude Code and other tools built on Anthropic models are responsible for nearly a third of all proposed code changes.¹ Agentic coding tools are AI tools that draft and propose code changes themselves, not just autocomplete a line.

That matters for the competitive picture because it measures adoption inside a workflow. A model that sits behind a third of proposed changes is part of how teams ship software, not a tool someone tried once. It also places Anthropic's models at the front of a use case that turns model quality directly into paid usage.

The note is the newest fact in the set, dated Oct 28, 2026.¹ So this is a current read on who leads in coding, and the obvious question for the next round of notes is whether that lead holds or narrows.

Value and trust: OpenAI's two soft spots

Coding share does not stand on its own. An industry research note found that Anthropic's subscription plans offer better value than OpenAI's for Opus 5.5.³ When a developer compares two subscriptions side by side, a value gap is one of the easiest reasons to switch.

Trust is the second front. OpenAI faces scrutiny over recent safety incidents involving rogue agents and leaks about its models.² For a buyer handing an AI agent access to code or systems, incidents with rogue agents are exactly the kind of headline that slows a decision.

Put together, the two notes describe a contest where Anthropic is helped by usage and value while OpenAI has to answer questions on safety.^1,2,3 Neither soft spot is permanent. Pricing can be changed, and scrutiny can fade if the response holds up. These are the two places where the picture could shift fastest.

OpenAI's counter-moves, and the challengers behind it

OpenAI is not standing still. It and Anthropic are competing to publish foundational theoretical results in computational complexity using AI.⁸ A research race like this is about reputation: it shows which lab's models can do frontier-level reasoning, and that reputation feeds recruiting and enterprise credibility.

On the product side, OpenAI's Decisions API sits in what the notes call the decision model market.⁴ That market already has challengers. TypeSafe AI's Jev model competes directly with OpenAI's Decisions API.⁴

Liquid AI has also stepped in, releasing D1 Ultra, a decision model that ranks highly on public benchmarks.⁵ Ranking well on public benchmarks gets a new model noticed. Turning that into paying customers is a separate, harder step.

So the competitive map has two layers. At the top, the large labs fight over coding, research and pricing. Below them, smaller labs are going after a newer product category where OpenAI has a product but no lock on the market.^4,5

The rival above: Google

Anthropic's gains do not come in a two-player race. It is one of Google's biggest AI competitors, and it competes directly with Google DeepMind.^6,7

That puts Anthropic's lead in a harder setting. Leading in agentic coding against OpenAI is one contest.¹ Holding that lead while Google DeepMind is also in the race is another.⁷ Leadership in a fast-moving field tends to draw a response, and Google is named as a direct competitor in the notes.

For readers mapping who is gaining and who is losing, the notes place Anthropic at the centre of two rivalries at once, OpenAI on one side and Google on the other.^8,6 That is a sign of strength, and also the main source of risk to the read.

What to watch next

The cleanest signal is coding share. If later notes still show tools on Anthropic's models responsible for around a third of proposed code changes, the usage lead looks durable.¹ If that share narrows, the next question is who took it.

Second, watch the value comparison. Anthropic's subscription plans were found to offer better value than OpenAI's for Opus 5.5.³ A pricing response from OpenAI would test how much of Anthropic's lead rests on value rather than on model quality.

Third, watch how the safety scrutiny on OpenAI develops.² A clear response would take away one of the arguments working against it with cautious buyers.

Finally, watch the decision model challengers. TypeSafe AI's Jev and Liquid AI's D1 Ultra show the category is open.^4,5 If either one reports customers to match its benchmark results, the list of closed-weight labs worth tracking gets longer, and that is useful for anyone weighing who supplies the next layer of AI products.