Google Releases Gemini 4 Hydrogen to All Users, Including Free App Users, Without Early-Access Testing Phase
Image to help understand the article
Google released Gemini 4 Hydrogen on Nov. 12, 2026, making its latest artificial intelligence model available immediately to all users, including those on the free Gemini app tier. The launch bypassed the two-week Fairwind Program that preceded the broader release of its earlier Gemini 4 Argon model, signaling a different rollout strategy for the new system.
Immediate Access Across Google's AI Platforms
The announcement, titled "Build with Gemini 4 Hydrogen," was published under the name of Koray Kavukcuoglu, Google's vice president and chief AI architect. Google positioned Hydrogen alongside Gemini 4 Carbon, Gemini 4 Argon and Gemini Decision within its AI product lineup.
The model is available through Google AI Studio and Android Studio, as well as Gemini for business. Users of Google Antigravity can access it with paid Google AI Pro or Ultra subscriptions. Unlike Argon, which initially went through a two-week Fairwind Program, Hydrogen was opened to everyone on launch day, including free Gemini app users.
According to Google DeepMind's model card, the API model ID is Gemini 4 Hydrogen. It supports a unified 1-million-token context window. Standard synchronous output supports up to 64,000 tokens, while background long-form decoding can extend output to 1 million tokens. Its stated knowledge cutoff is March 2026, although Google cautions that some areas may contain older information.
Benchmark Results Show Strengths Against OpenAI and Anthropic
Google's published benchmarks show Hydrogen outperforming several competing models in selected categories. In the SWE-sweep agentic coding evaluation, it scored 13.5%, compared with 3.1% for Anthropic's Claude Sonnet 5.5. On Agents' Last Exam v1.1, which evaluates knowledge-intensive work, Hydrogen achieved 71.6%, ahead of OpenAI's GPT-6.1 Terra at 69.8%. It also scored 58.8% on the HLE-Diamond reasoning evaluation, compared with 53.3% for Claude Sonnet 5.5.
On Terminal-Bench 4.1, Hydrogen recorded 57%, while Google's Gemini 3.8 Flash scored 18.2%. However, Hydrogen did not lead across every category. It trailed GPT-6.1 Terra and Claude Sonnet 5.5 in science and mathematics benchmarks, computer-use evaluations and Terminal-Bench 4.1, although Google described the gaps in those comparisons as relatively small.
Google also reported improvements in enterprise testing. For complex data tasks, Hydrogen operated up to 2.8 times faster while using up to 23% fewer tokens. The company said instances in which AI agents acted beyond the scope of their instructions fell by as much as 81%.
In the Artificial Analysis Intelligence Index, Hydrogen scored 51 points at the standard high reasoning setting, compared with 50 for Gemini 4 Argon. At the newer xhigh reasoning level, its score increased to 54.
Independent Evaluation Places Hydrogen Sixth Overall
Independent testing by Vals AI produced a more measured picture of Hydrogen's performance. The model scored 68.49%, with a margin of plus or minus 0.99 percentage points, placing sixth among 37 models in the overall Vals Index. The reported cost per evaluation run was $9.838.
Its strongest result came in tax-related AI tasks. Hydrogen ranked first among 85 models in the Tax Agent Bench with a score of 52.30%, ahead of Gemini 4 Carbon at 51.78% and Claude Fable 5.1 at 50.75%. It ranked second among 40 models in the Vals Finance Index, scoring 66.14%, and ninth among 38 models in the Vals Legal Index, with 73.63%.
API Pricing Favors Longer Requests
Hydrogen's standard API pricing is $1.50 per million input tokens without caching, $0.075 per million cached input tokens and $7.50 per million output tokens. Batch, Flex and Contributor tiers receive a 50% discount. Google is not offering a special launch discount, and explicit caching is unavailable at launch.
Against other leading models, Hydrogen occupies a middle position on standard token pricing. GPT-6.1 Terra costs $1 per million input tokens and $5 per million output tokens, while Claude Sonnet 5.5 costs $2 and $10, respectively. Hydrogen is therefore 50% more expensive than Terra but 25% cheaper than Sonnet 5.5 for requests with fewer than 272,000 prompt tokens, assuming equal token volumes. For longer requests exceeding that threshold, Hydrogen becomes the least expensive of the three.
Google's pricing also makes Hydrogen a more economical option than its sibling Argon for extended workloads. Argon's introductory rate of $2 per million input tokens and $10 per million output tokens is scheduled to end Nov. 30, 2026, when prices rise to $4 and $20, respectively. By comparison, Hydrogen's rates remain at $1.50 and $7.50.
At its standard input and output rates, Hydrogen costs 15 times as much as lower-priced models including Claude Haiku 5.5, GPT-6 Luna and Meta Muse Ember. The pricing structure therefore makes the model's relative value dependent on workload length, performance requirements and token consumption.
Google Acknowledges Remaining AI Limitations
Google's model card also identifies limitations that remain common across generative AI systems. Hydrogen can produce hallucinations, and the company lists resistance to jailbreak attacks as an area where further improvements are ongoing.
The launch gives developers, businesses and individual users immediate access to Google's newest model, while its benchmark results suggest particular strengths in tax-related tasks, knowledge work and selected reasoning and coding evaluations. Its independent overall ranking and remaining weaknesses, however, show that strong results in individual categories do not necessarily translate into leadership across every AI application.
Comments
Post a Comment