Google Tests Internal Gemini 4 "Carbon" for Coding; Staff Say It Rivals Claude Opus 5.5

What happened

On October 9, Business Insider reported, citing internal documents and employee interviews, that Google has deployed a new Gemini 4 checkpoint codenamed "Carbon" to Jetski, its internal coding platform (associated with Antigravity), for employee testing. One tester said Carbon's coding performance is comparable to Anthropic's Claude Opus 5.5; another gave it positive marks as well, though the first cautioned that broader testing is still needed.

Key facts

1. Internal codename Carbon:reportedly a new checkpoint in the Gemini 4 family, currently tested only on the internal Jetski platform and not publicly released.
2. Employee feedback:one tester likened its coding ability to Anthropic's Claude Opus 5.5; another was similarly positive but noted broader testing is needed.
3. Release form undecided:Business Insider could not establish whether Carbon will ship as an Argon update or a separate model; Google may never release it publicly. Google declined to comment and published no benchmark results for Carbon.
4. Multiple versions in parallel:internal documents show Google evaluating several Gemini 4 versions under the Argon, Barium and Carbon codenames; Barium-B was reportedly chosen to become the public Argon.
5. Background:Google unveiled Gemini 4 Argon on September 30, first to cybersecurity partners in its Fairwind Program; staff previously said early Argon versions lagged rivals on some coding tasks.