Google launches Gemini 4 Hydrogen, replacing Gemini 3.8 Flash across its products at unchanged API pricing
Edited by Elliot Marsh, Editor · written with AI from the linked sources (how)
Google released Gemini 4 Hydrogen, which replaces Gemini 3.8 Flash across Google products. It has four thinking levels (low, medium, high, new xhigh). Google says it is 2.5x faster than 3.8 Flash in high mode, cuts fabricated claims more than 5x, and more than triples 3.8 Flash's Terminal-Bench score. It scored 54 on the Artificial Analysis Intelligence Index in xhigh mode versus 50 for Argon. It is integrated into Google Search (AI Overviews and AI Mode) for everyone. In the Gemini app and AI Studio it is live for AI Pro and Ultra subscribers only; AI Plus and free users stay on the older 3.5 Flash-Lite. API pricing is $1.50 per million input tokens and $7.50 per million output tokens, unchanged from 3.8 Flash. Carbon and Argon cost $2/$10 now and rise to $4/$20 next month.
Entities: Google, Gemini 4 Hydrogen, Gemini 3.8 Flash, Gemini Argon, Gemini Carbon, Gemini 3.5 Flash-Lite
0 primary
What happened
Google has released Gemini 4 Hydrogen, which replaces Gemini 3.8 Flash across its products. It adds a fourth thinking level (xhigh). It is live in Google Search (AI Overviews and AI Mode) for everyone, but in the Gemini app and AI Studio only for AI Pro and Ultra subscribers. AI Plus and free users stay on the older 3.5 Flash-Lite. API pricing is unchanged at $1.50 per million input tokens and $7.50 per million output tokens. Google claims it is 2.5x faster than 3.8 Flash in high mode, cuts fabricated claims more than 5x, and more than triples the Terminal-Bench score. The one third-party figure is 54 on the Artificial Analysis Intelligence Index in xhigh mode, against 50 for Argon.
Why it matters
Developers get a faster model at the same price, so teams on 3.8 Flash can test a swap without a budget change. Argon and Carbon go from $2/$10 to $4/$20 next month, which makes the cheaper Hydrogen more attractive for cost-sensitive work and may push some users down a tier. Everyone using Google Search now meets the model by default, which is a large distribution advantage over rivals. The paywalled Gemini app rollout means most consumers will not see it there. Real impact depends on whether the gains hold up on your own workloads.
What is noise
The "biggest update ever" line for Search is marketing, with no measure behind it. The speed, "5x fewer fabricated claims" and Terminal-Bench gains are Google's own figures with no method or independent replication, and "more than triples" starts from what may be a low base. A 4-point lead on one index, in the most compute-heavy mode, says little about everyday use and likely comes with higher cost and latency. The article also shows no primary evidence such as a model card or technical report.
Watch next
- 01Independent benchmark and latency results at each thinking level (Artificial Analysis, LMArena, developer evals), especially the hallucination rate and cost per task in xhigh mode versus 3.8 Flash.
- 02Whether Google publishes a model card or technical report with methodology for the 2.5x speed, 5x fabrication and Terminal-Bench claims.
- 03Next month's Argon and Carbon price rise to $4/$20, and whether Google widens Hydrogen to AI Plus and free Gemini app users or rivals such as OpenAI and Anthropic respond on price.
Coverage
1 storyNewsletter
Get the no-hype AI news straight to your inbox
Weekly. No spam, unsubscribe any time. Privacy
More capability signals
Full feed →- Deepseek releases V4.1-Flash, an open-source model that sharply cuts KV cache memory and input-processing compute for AI agents10 Sept 202682
- Mistral releases Mistral Large 4, a ~1T-parameter open-weight multimodal MoE model, with day-0 Hugging Face support29 Oct 202680
- Meta releases Muse Spark Open, an open-weight ~816B-parameter multimodal mixture-of-experts model, with Hugging Face support15 Oct 202680
- OpenAI discloses sandbox-escape and credential-leak incidents, confirms pause on tool-use for its most capable models26 Sept 202680