Is Google finally catching up with Anthropic and OpenAI?

Google has stepped up its frontier-AI challenge, opening Gemini 4 Carbon to wider use on November 10 and launching the faster Hydrogen on November 12. Published tests put Carbon ahead in some coding and business workflows, while other results favor Anthropic and OpenAI.

What did Google release?

Google positions Carbon as its strongest system for demanding, extended work in coding, research and cybersecurity. Hydrogen serves a different role: faster everyday execution, with a one-million-token context window. It supplements Carbon rather than replacing it as the deepest-reasoning option.

Where does Carbon lead?

On CursorBench 4.0, which uses multi-file coding tasks from real sessions, Carbon at High effort scores 60.4%, narrowly above Fable 5.2 at Max effort on 60.0%. Recorded costs are $9.22 versus $32.79 per task. Those are results for these specific configurations, not identical reasoning budgets.

On Zapier’s AutomationBench, Carbon High scores 53.88% at $2.12 per task, against 51.45% and $4.64 for Fable 5.2 Max with default fallbacks. The benchmark grades completed business workflows, not conversational polish.

Where do Anthropic and OpenAI stay ahead?

Anthropic’s November 9 release introduced Fable 5.2 and Mythos 5.2: the same underlying system with different safeguards and access. Fable is broadly available; Mythos access is restricted through verification programs.

In Anthropic’s own FrontierSWE v2 evaluation, Fable scores 69.9%, GPT-6 Astra 65.5% and Carbon 61.2%. These vendor-run results favor different leaders from the independent workflow tests. They do not establish a single overall winner.

OpenAI identifies GPT-6 Astra as its flagship; GPT-6.1 Sol balances capability and cost. ChatGPT Work and Codex offer Astra, while app availability and settings vary. Comparing the named systems is more precise than treating every ChatGPT conversation as the same product configuration.

Is price part of Google’s challenge?

Yes. Carbon’s introductory API rates are $2 per million input tokens and $10 per million output tokens. Astra’s listed rates are $10 and $50. Token prices alone do not measure the cost of completing a task; reasoning length and tool use also matter.

So, is Google catching up?

On these published results, Google is already competitive—and ahead on some practical tests. The narrower conclusion is stronger than a blanket victory claim: coding efficiency and business execution are credible strengths, while other engineering evaluations still favor its rivals.