We gave GPT-6.2 Astra and Sol the same cube. Both solved it in the Tabbit browser with Tabbit CLI.
Astra · 3m 41.6s · $0.80 · 921,246 tokens
Sol · 3m 50s · $0.22 · 714,651 tokens
Would you wait 8 more seconds to spend 72% less?
Qwen4-Max is now free to use in Tabbit.
Got a long report open? Try asking, “What’s this chart actually telling me?”
Qwen4-Max can read the chart and the text around it, with a 1M-token context window for when the answer takes more than a page.
2
219
Replying to @Alibaba_Qwen
Qwen4-Max is now free to use in Tabbit.
Got a long report open? Try asking, “What’s this chart actually telling me?”
Qwen4-Max can read the chart and the text around it, with a 1M-token context window for when the answer takes more than a page.
284
We tried translating an 1859 math paper in Tabbit. The equations stayed intact.
Tabbit 1.21 now preserves the structure of math formulas when translating PDFs. Here’s what that looks like ↓
1
3
311
Replying to @Google @GoogleDeepMind
We put Gemini 4 Carbon and Claude Opus 5.5 head-to-head on Chinese writing in Tabbit.
EssayBench results: Gemini 4 Carbon: 91.1/100 Claude Opus 5.5: 87.6/100
Same prompt, side-by-side outputs below.
1
273
Replying to @claudeai
We asked Claude Fable 5.2 and GPT-6 Astra to draw Buddy in Canva using Tabbit CLI.
GPT-6 Astra: 8m 54s · $3.94
Claude Fable 5.2: 6m 35s · $2.83
Both on high reasoning, no fast mode. Costs estimated at API rates (USD).
15
We asked Claude Fable 5.2 and GPT-6 Astra to draw Buddy in Canva using Tabbit CLI.
GPT-6 Astra: 8m 54s · $3.94
Claude Fable 5.2: 6m 35s · $2.83
Both on high reasoning, no fast mode. Costs estimated at API rates (USD).
Which one would Buddy hire? 🦄
7
15
981
Tabbit Model Chart Vol. 11
Hy4 debuts at #11. Gemini-4-Argon stays ahead of Sonnet 5.5 and holds #1. The top five remain unchanged, which points to strong use-case stickiness.
We've received a lot of user feedback asking whether deep thinking can be adjusted or turned off. To meet different needs, Tabbit will soon support customizable model thinking intensity, bringing a more flexible balance between response speed, cost, and depth of reasoning.
1
1
210
For users who may enjoy more hands-on control, we've integrated the DeepSeek Harness (DSH) environment into Tabbit.
Install DSH in one click by entering tabbit://local-tools in the Tabbit address bar. Once installed, DSH runs directly inside Tabbit, with no command-line setup required.
DSH usage isn't covered by your Tabbit allowance. Tabbit will sync your DeepSeek API key from any other local DSH install, or you can add keys manually in Settings.
2
4
705














