cd /news/large-language-models/new-models-tested-gemini-4-hydrogen-… · home › topics › large-language-models › article
[ARTICLE · art-172233] src=metergraph.dev ↗ pub= topic=large-language-models verified=true sentiment=· neutral

New models tested: Gemini 4 Hydrogen, Qwen4-Max, and GPT-6.2

Independent five-workload testing found GPT-6.2 Sol nearly matched GPT-6.1 Sol on mean quality (0.913 versus 0.919) while cutting latency from 25.3 s to 16.4 s at a similar $0.0342 per task, and GPT-6.2 Astra led on coding at 0.928 but cost $0.1278 per task. Gemini 4 Hydrogen beat Gemini 4 Argon on mean quality (0.859 versus 0.844) but was slower (53.8 s versus 40.7 s) and cost more ($0.0778 versus $0.0430 per task), while Qwen4-Max reached 0.877 mean quality at $0.0428 per task but averaged 102.5 s of latency, with coding tasks taking about five minutes. The tester recommended starting with GPT-6.2 Sol, using Hydrogen for design tasks, and noted Qwen4-Max is no longer open-weight.

recommendation

Route coding-heavy work to GPT-6.2 Astra only when its 0.928 coding score justifies $0.1278 per task, and default to GPT-6.2 Sol at $0.0342 for general workloads.

more in playbook: Controlling LLM cost
by read6 min views5 publishedNov 19, 2026
New models tested: Gemini 4 Hydrogen, Qwen4-Max, and GPT-6.2
Image: source

Summary #

I tested Gemini 4 Hydrogen, Qwen4-Max, and GPT-6.2 Astra and Sol across my standard five workloads.

Can you guess whether Hydrogen is above or below Argon? I expected a faster version of Argon based on the name, but it was actually smarter, slower, and more expensive. Google is doing a great job of confusing us.

GPT-6.2 Sol is a faster version of GPT-6.1 Sol. GPT-6.2 Astra led on coding, but like the new Fable, it comes at a steep price.

Qwen4-Max is still behind the latest Sol, but it narrowed the gap. It is competitive on quality and cost, though it is still significantly slower.

GPT-6.2 Sol and Astra #

GPT-6.2 Sol was nearly as strong as GPT-6.1 Sol overall and substantially faster. Astra did best on coding among the models in this run, but fell behind on extraction and cost several times more per task. Scores vary a little across workloads, but the two Sol models show similar strengths overall.

Model Mean quality Cost / task Latency
GPT-6.2 Sol 0.913 $0.0342 16.4 s
GPT-6.1 Sol 0.919 $0.0339 25.3 s
GPT-6.2 Astra 0.904 $0.1278 22.9 s
GPT-6 Astra 0.900 $0.1719 28.0 s

Quality by workload #

Model Design Coding Extraction Matching Research
GPT-6.2 Sol 0.726 0.905 0.973 0.963 1.000
GPT-6.1 Sol 0.746 0.905 0.991 0.982 0.971
GPT-6.2 Astra 0.719 0.928 0.911 0.960 1.000
GPT-6 Astra 0.643 0.907 0.991 0.957 1.000

My takeaway: Start with GPT-6.2 Sol; consider Astra where coding quality matters enough to justify the cost.

Gemini 4 Hydrogen and Argon #

Hydrogen scored higher than Argon overall in this run, with stronger design and research results, but it was slower and more expensive.

Model Mean quality Cost / task Latency
Gemini 4 Hydrogen 0.859 $0.0778 53.8 s
Gemini 4 Argon 0.844 $0.0430 40.7 s
Gemini 3.8 Flash 0.762 $0.0439 53.5 s

Quality by workload #

Model Design Coding Extraction Matching Research
Gemini 4 Hydrogen 0.543 0.811 0.977 0.968 0.998
Gemini 4 Argon 0.453 0.825 0.996 0.991 0.953
Gemini 3.8 Flash 0.167 0.702 0.995 0.982 0.963

My takeaway: Use Hydrogen for design tasks, but otherwise I would prefer Argon.

Qwen4-Max #

Qwen4-Max scored higher overall than Qwen 3.8 Max. Its workload scores were close to GPT-6 Sol’s, but it was much slower, particularly on coding.

Model Mean quality Cost / task Latency
Qwen4-Max 0.877 $0.0428 102.5 s
Qwen 3.8 Max 0.830 $0.0933 74.3 s
GPT-6 Sol 0.881 $0.0291 12.2 s
GPT-6.2 Sol 0.913 $0.0342 16.4 s

Quality by workload #

Model Design Coding Extraction Matching Research
Qwen4-Max 0.623 0.859 0.961 0.976 0.965
Qwen 3.8 Max 0.503 0.742 0.936 0.969 1.000
GPT-6 Sol 0.648 0.877 0.945 0.965 0.971
GPT-6.2 Sol 0.726 0.905 0.973 0.963 1.000

My takeaway: A capable model, but coding tasks took about five minutes on average. It is unfortunate that it is not open-weight anymore, which was its main selling point in the past.

Testing results #

Higher quality is better. The charts include only results meeting the coverage threshold.

Design quality

Cost per task

Five-workload mean; lower is better.

Response time

Five-workload mean in seconds; lower is better.

Full benchmark data #

Each row is a model and a workload. Quality is on a 0–1 scale. Costs and latency are means of successfully measured requests for that workload, excluding judge and setup costs. Under-covered scores are shown for inspection, not ranked.

Download the full results JSON · Download results as CSV

Model Workload Quality Cost / task Latency
GPT-6.1 Sol Design 0.746 $0.0630 74.5 s
GPT-6.1 Sol Coding 0.905 $0.0790 32.0 s
GPT-6.1 Sol Extraction 0.991 $0.0110 7.7 s
GPT-6.1 Sol Matching 0.982 $0.0022 2.4 s
GPT-6.1 Sol Research 0.971 $0.0145 9.8 s
GPT-6.2 Sol Design 0.726 $0.0589 28.1 s
GPT-6.2 Sol Coding 0.905 $0.0865 40.8 s
GPT-6.2 Sol Extraction 0.973 $0.0111 8.1 s
GPT-6.2 Sol Matching 0.963 $0.0021 1.5 s
GPT-6.2 Sol Research 1.000 $0.0123 3.8 s
GPT-6.2 Astra Design 0.719 $0.2209 55.5 s
GPT-6.2 Astra Coding 0.928 $0.3024 34.2 s
GPT-6.2 Astra Extraction 0.911 $0.0572 12.4 s
GPT-6.2 Astra Matching 0.960 $0.0090 3.5 s
GPT-6.2 Astra Research 1.000 $0.0497 8.9 s
Gemini 4 Argon Design 0.453 $0.0803 79.7 s
Gemini 4 Argon Coding 0.825 $0.1056 75.7 s
Gemini 4 Argon Extraction 0.996 $0.0130 20.9 s
Gemini 4 Argon Matching 0.991 $0.0022 9.2 s
Gemini 4 Argon Research 0.953 $0.0138 17.9 s
Gemini 4 Hydrogen Design 0.543 $0.1466 104.8 s
Gemini 4 Hydrogen Coding 0.811 $0.1942 131.1 s
Gemini 4 Hydrogen Extraction 0.977 $0.0223 16.6 s
Gemini 4 Hydrogen Matching 0.968 $0.0067 5.5 s
Gemini 4 Hydrogen Research 0.998 $0.0191 10.9 s
Qwen4-Max Design 0.623 $0.0565 146.9 s
Qwen4-Max Coding 0.859 $0.1295 302.3 s
Qwen4-Max Extraction 0.961 $0.0119 29.8 s
Qwen4-Max Matching 0.976 $0.0021 5.1 s
Qwen4-Max Research 0.965 $0.0140 28.4 s

Methodology #

Five workloads, eight cases per workload, two repetitions. Sonnet 5.5 graded saved responses using the published criteria and case-specific design rubrics. Quality scores reflect rubric performance, not task pass rates. Overall quality, cost and latency give equal weight to each workload; this can differ from averaging all successful requests together.

Incomplete outputs and explicit format failures score zero; provider failures are excluded. A model/workload result needs at least 80% coverage to be included in comparisons. Sampling and judge variation can affect comparisons.

── more in #large-language-models 4 stories · sorted by recency
── more on @gemini 4 hydrogen 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/new-models-tested-ge…] indexed:0 read:6min 2026-11-19 · —