Ternary Bonsai
Ranked 2 of 4 quantized language models by recommenders
Compact 1.58-bit model people praise for accuracy per byte and fast on-device use. Critics say it hallucinates often and is weak at coding and logic.
Compact 1.58-bit model people praise for accuracy per byte and fast on-device use. Critics say it hallucinates often and is weak at coding and logic.
22 people gave an opinion
- recommend it
- 14
- criticize it
- 8
- approval share
- 64%
Recommended in 3 of 13 months
- November 2025: not recommended
- December 2025: not recommended
- January 2026: not recommended
- February 2026: not recommended
- March 2026: not recommended
- April 2026: recommended
- May 2026: recommended
- June 2026: not recommended
- July 2026: recommended
- August 2026: not recommended
- September 2026: not recommended
- October 2026: not recommended
- November 2026: not recommended (month in progress)
Why it doesn’t carry the badge
- 14 people recommend itNeeds 10 or more, met.
- 1.7 recommenders for every criticNeeds 2 to 1 or better, not met.
- Recommended in 3 of 13 monthsNeeds 4 or more, not met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Strong accuracy for its size
“Ternary-Bonsai-1.7B achieved 65.1% from 462 MiB”
Comment from 21 Apr 2026 in “Ternary Bonsai: Top Intelligence at 1.58 Bits” “Essentially seems to be a very good quantization of that model”
Comment from 21 Apr 2026 in “Ternary Bonsai: Top Intelligence at 1.58 Bits” “good-per-byte-but-not-in-absolute-terms quantization”
Comment from 21 Apr 2026 in “Ternary Bonsai: Top Intelligence at 1.58 Bits” Runs well for on-device AI
“This is currently the best model for on-device AI and has been used in our three apps.”
Comment from 16 Apr 2026 in “Ternary Bonsai: Top Intelligence at 1.58 Bits” “it's great progress toward local AI, which I believe will be common in the near future”
Comment from 11 May 2026 in “Local AI needs to be the norm” “Vram usage was consistently at ~7GiB.”
Comment from 21 Apr 2026 in “Ternary Bonsai: Top Intelligence at 1.58 Bits” Fast tokens and first response
“it seems to get higher TPS than most other similarly sized models on my hardware”
Comment from 14 Jul 2026 in “Bonsai 27B (1-bit LLM): The First 27B-Class Model to Run on a Phone” “time to first token is incredibly low”
Comment from 21 Apr 2026 in “Ternary Bonsai: Top Intelligence at 1.58 Bits”
What people warn about
Each mark is one comment
Weak at coding tasks
“works great for basic conversational tasks and not good for one shot coding tasks”
Comment from 21 Apr 2026 in “Ternary Bonsai: Top Intelligence at 1.58 Bits” “bonsai is basically trash on real agentic coding work”
Comment from 15 Oct 2026 in “LFM2.5-24B-A2B: Built for fast on-device agents” Benchmarks look cherry-picked
“Independent testing of prismml suggest quite a capability drop off outside of their cherry picked benchmarks”
Comment from 28 Jul 2026 in “Neutrino-1 8B” “the comparisons read as a bit disingenuous with the other models at 16 bit weights”
Comment from 21 Apr 2026 in “Ternary Bonsai: Top Intelligence at 1.58 Bits” Hallucinates answers often
“much like the bonsai tenary one it hallucinates knowledge quite aggressively”
Comment from 4 Aug 2026 in “Show HN: Maple-Preview – ternary 20B MoE running at 120 tok/s on a iPhone” “hallucinates in pretty much every answer”
Comment from 21 Apr 2026 in “Ternary Bonsai: Top Intelligence at 1.58 Bits” Weak logic and literal answers
“Fails the car wash test (basic logic seems to be weak in general)”
Comment from 21 Apr 2026 in “Ternary Bonsai: Top Intelligence at 1.58 Bits” “This model tends to be annoyingly literal.”
Comment from 21 Apr 2026 in “Ternary Bonsai: Top Intelligence at 1.58 Bits”
Other quantized language models people recommend
The three most recommended other quantized language models, with the same counts for each, so you can compare like for like. The tick on each bar marks the 2 to 1 line the badge needs.
| Product | Recommend | Criticize | Approval | Months | Owned | Badge |
|---|---|---|---|---|---|---|
| Microsoft BitNet | 16 | 9 | 64% | 4 | — | — |
| Ternary BonsaiThis page | 14 | 8 | 64% | 3 | — | — |
| Maple-Preview | 6 | 1 | 86% | 1 | — | — |
| BitNet b1.58 2B4T | 5 | 1 | 83% | 4 | — | — |
What people said
22 of the 25 comments that mention it, newest first. Each links to the original comment.
“bonsai is basically trash on real agentic coding work”
“much more recent, but worse models like 6.7GB on disk size ternary bonsai”
“much like the bonsai tenary one it hallucinates knowledge quite aggressively”
“I really had high hopes for the larger Ternary Bonsai and it feels like there is scope to improve”
“Independent testing of prismml suggest quite a capability drop off outside of their cherry picked benchmarks”
“it seems to get higher TPS than most other similarly sized models on my hardware”
“it's great progress toward local AI, which I believe will be common in the near future”
“the comparisons read as a bit disingenuous with the other models at 16 bit weights”
“good-per-byte-but-not-in-absolute-terms quantization”
“time to first token is incredibly low”
“Essentially seems to be a very good quantization of that model”
“works great for basic conversational tasks and not good for one shot coding tasks”
“Fails the car wash test (basic logic seems to be weak in general)”
“hallucinates in pretty much every answer”
“Vram usage was consistently at ~7GiB.”
“This model tends to be annoyingly literal.”
“Wow, if this is true, I am extremely impressed and excited!”
“Ternary-Bonsai-4B is the accuracy-per-byte winner above 1 GiB.”
“in accuracy-per-byte, bonsai is the clear winner”
“Ternary-Bonsai-1.7B achieved 65.1% from 462 MiB”
“the big advantage of 1.58 bits is there are no multiplications at inference time”
“This is currently the best model for on-device AI and has been used in our three apps.”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.