Research · Head-to-head

Inkling-Small vs Bonsai 27B

Inkling-Small (free, AI Score 8.8/10) vs Bonsai 27B (free, AI Score 8.5/10). Side-by-side pricing, features, pros and cons, and which to pick.

The verdict

Pick Inkling-Small if…

Both are credible in this slot.

Try Inkling-Small →

Pick Bonsai 27B if…

→your primary use case is developers and researchers building offline or privacy-sensitive agentic apps who need a multimodal model small enough to run locally on a phone or laptop.
→you need: development, agents, research

Try Bonsai 27B →

Spec	Inkling-Small	Bonsai 27B
Category	Research	Research
Pricing model	free	free
Headline pricing	Free — open weights download; Tinker Playground access	Free — open-source weights under Apache 2.0
Free tier	Yes — the weights themselves are free to download and run. Compute is the real cost.	Entirely free and open-source — weights released under Apache 2.0 with no paid tier from PrismML.
AI Score	8.8/10	8.5/10
Best for	—	Developers and researchers building offline or privacy-sensitive agentic apps who need a multimodal model small enough to run locally on a phone or laptop.
Editor's pick	✓ Yes	✓ Yes
Use cases	—	development agents research
Date added	2026-07-31	2026-07-15

Research · free

✓Open weights on Hugging Face, so it can run fully offline for sensitive audio, documents, or code
✓12B active parameters keeps per-token inference cost near a mid-size dense model despite the 276B total
✓Audio and vision are native to the model rather than a separate encoder stage
✓1M-token context handles whole codebases or long document sets in one pass
✓Hosted Tinker Playground path for teams that want to evaluate before committing hardware

×"Free weights" is misleading on cost — serving 276B parameters requires enough VRAM to put local deployment out of reach for individuals and small teams
×The claim that it outperforms the larger Inkling comes from the lab's own launch benchmarks; independent third-party evaluations aren't in yet
×Released July 30, 2026, so tooling, quantizations, and community fine-tunes are still thin compared to established open-weights families
×It's a raw model, not a product — no chat app, no agent harness, no support contract unless you build or buy one

Research · free

✓27B-class model small enough to run on a phone via ternary/1-bit weights — a genuinely new footprint for this size class
✓Apache 2.0 license permits commercial use, fine-tuning, and redistribution with no strings attached
✓Fully local inference keeps data on-device, a real advantage for privacy-sensitive and offline apps
✓Multimodal rather than text-only, broadening what on-device agentic workflows can do
✓Free — you only pay for your own compute

×'Near full-precision' claims at 1-bit/ternary are the vendor's own and need independent benchmarking before you trust them
×Running a 27B model on a phone still taxes RAM, thermals, and battery — real-world throughput on older devices is unproven
×Self-hosting means you handle deployment, quantization tooling, and updates; there's no managed API to fall back on
×Brand-new (launched July 14, 2026), so tooling, community quants, and long-term support are still immature

Updated 2026-07-31. Spec data sourced from official product pages and tracked in our public directory at /tools.