Google's Varun Mohan Says Carbon's Time to First Token Is Now 1.5x Faster in Antigravity
Time to first token (TTFT) for Carbon is now 1.5x faster on Antigravity. Will follow soon with Argon too. Lots of work went into this so you all could have a faster experience. More to come 馃憖
Context
Faster first responses make coding agents feel more usable, and Mohan says the same speedup is coming to Argon.
Antigravity is Google's agent based coding environment, led by Varun Mohan. Carbon and Argon are model options available in the tool. Time to first token measures how long a model takes to begin answering.
Deep dive
Varun Mohan posted that time to first token for Carbon is now 1.5x faster on Antigravity, and that the same improvement will follow soon for Argon. He wrote that a lot of work went into the change so users could have a faster experience, and hinted that more is coming.
Time to first token is the delay between sending a request and seeing the first part of the model's answer. In coding tools, where agents make many calls in a row, shorter delays add up and make the product feel more responsive, even when total generation time is unchanged.
The post does not give absolute latency numbers, the baseline being compared, or whether the gain comes from serving infrastructure, caching or model changes. It also does not say whether it applies to all users or plans.
The update comes a day after Mohan posted that Antigravity turned one year old and reset usage quotas for everyone across all models, which is covered in a separate story. The post does not spell out Carbon's full model name, though Artificial Analysis listed a Gemini 4 Carbon model at 60 on its Intelligence Index in its post about GPT 6.2.
What to expect next
Watch for the Argon speedup Mohan promised and for any published latency figures that let users compare models inside Antigravity. Developers will judge the change by how quickly long agent runs complete in practice. Google may also announce further performance or pricing updates as competition with OpenAI's Codex and Anthropic's Claude Code continues. Users can also compare the change against Codex and Claude Code response times in their own work.
