A Year On, Qwen3 Still Holds the Price/Quality Throne in Our Model Battle Test
In our LLM battle test, Qwen3-235B from July 2025 again leads on price/quality. Over the year Gemini gained 40 points and DeepSeek V4 Flash turned out to be underrated by 6 points, but neither dislodged Qwen. We also tested the models' generation speed.
AI-processed from Habr AI; edited by Hamidun News
I put four LLMs into a single batch to check whether the small Gemma really beat the large one in cross-session tests. The results turned out to be far more interesting than expected.
Head-to-head: the two Gemmas did not separate
In a fair head-to-head, the surprising result of the cross-session test evaporated: both Gemma models came out level, with no difference between them. But that was only the beginning. DeepSeek V4 Flash, which I had rated at 83 points, scored 89 this time, exactly 6 points higher. The model had been underrated, and that became the main finding of the batch.
Overrating one model can lead to underrating the whole hierarchy. That is why fair head-to-head comparisons in a single context remain the gold standard.
Qwen has held the throne for a year
Qwen3-235B-A22B-2507 (released July 21, 2025) once again took first place for price/quality. That was the July checkpoint, almost exactly a year ago. And it still has not been displaced by its competitors.
A lot happened over that year. Gemini jumped from 57 to 97 points, a gain of 40. I retested DeepSeek three times, each time with new results. New contenders appeared. But Qwen? It simply holds the throne.
- Gemini: +40 points over the year
- DeepSeek V4 Flash: underrated by 6 points
- Qwen3: still the best on price/quality
- MiniMax: got a lot of hype; solid in tests, but not revolutionary
- Eight new models from June: did not move the leader
New criteria and the MiniMax hype
The ranking update adds a new criterion: generation speed. It turns out that speed and quality do not always go hand in hand. A model can be fast but slower at learning from current data, or the other way around.
MiniMax deserves a separate mention. Everyone does praise it, and in capabilities it is close to Opus. But it came with a very active promotion campaign. In a fair test it delivers results worth attention, but not revolutionary enough to rewrite the hierarchy.
What this means
If you are choosing between quality and price, Qwen3-235B remains the best choice for most tasks. The other models are more specialized: Gemini for multimodality, DeepSeek for experimentation, MiniMax for those willing to pay more.
- Nov 1, 2026Google's StreamVibe video agent beats other models by 10.7% on response timing and quality
- Nov 1, 2026OpenAI Rewrites GPT Behavior Rules: Five Values and a Chain of Command
- Nov 1, 2026Voice bots built on LLMs turn to Jev, a fast classifier, for decisions in 200–500 ms
- Nov 1, 2026Google admits AI Mode cuts search revenue, shares fall 15%
Key Questions Answered
What are the key details on Head-to-head: the two Gemmas did not separate?
In a fair head-to-head, the surprising result of the cross-session test evaporated: both Gemma models came out level, with no difference between them. But that was only the beginning. DeepSeek V4 Flash, which I had rated at 83 points, scored 89 this time, exactly 6 points higher. The model had been underrated, and that became the main finding of the batch. Overrating one model can lead to underrating the whole hierarchy. That is why fair head-to-head comparisons in a single context remain the gold standard.
What are the key details on Qwen has held the throne for a year?
Qwen3-235B-A22B-2507 (released July 21, 2025) once again took first place for price/quality. That was the July checkpoint, almost exactly a year ago. And it still has not been displaced by its competitors. A lot happened over that year. Gemini jumped from 57 to 97 points, a gain of 40. I retested DeepSeek three times, each time with new results. New contenders appeared. But Qwen? It simply holds the throne. Gemini: +40 points over the year DeepSeek V4 Flash: underrated by 6 points Qwen3: still the best on price/quality MiniMax: got a lot of hype; solid in tests, but not rev
What are the key details on New criteria and the MiniMax hype?
The ranking update adds a new criterion: generation speed. It turns out that speed and quality do not always go hand in hand. A model can be fast but slower at learning from current data, or the other way around. MiniMax deserves a separate mention. Everyone does praise it, and in capabilities it is close to Opus. But it came with a very active promotion campaign. In a fair test it delivers results worth attention, but not revolutionary enough to rewrite the hierarchy.
Test Frontier LLMs & Autonomous Agents in One Workspace
Run Claude Opus 5.5, GPT-6, Gemini 3.1, and DeepSeek side-by-side in Nomad AI (nomadnet.ai) — unified model routing, deep research, and multimodal workflows under one subscription.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.