Tasks and inputs
Bolmo 1B is listed for Text generation, Text-to-text AI models. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Bolmo 1B is Ai2's small open byte-level language model, OLMo 2 1B converted to work on raw UTF-8 bytes. Apache-2.0, research use.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Open weights for self-hosting; no hosted API price. Hardware cost is yours.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Bolmo 1B is the cheap way to find out whether byte-level modelling helps your problem. It runs the same byte-level design as the larger Bolmo 7B, in a package small enough for a laptop GPU or a quick fine-tuning sweep.
Ai2 introduced the Bolmo family on 15 December 2025.
Bolmo is generally described as byteified Olmo 3. That holds for the 7B. The 1B starts from OLMo 2 1B instead (OLMo-2-0425-1B), via an intermediate Bolmo-1B-Stage1 checkpoint. If you need to know which backbone you are getting, the checkpoint's base-model field is the thing to trust: this is an OLMo 2 descendant.
Briefly:
Training follows the family recipe: freeze the transformer and train the new byte components, then unfreeze and train the whole model.
| Hub ID | allenai/Bolmo-1B |
| Parameters | ~1.5B (stored in FP32) |
| Language | English |
| Data | bolmo_mix |
| Dependencies | transformers 4.57.3+, xlstm==2.0.4, Python 3.11+ |
| Licence | Apache-2.0, research and education under Ai2 guidelines |
Remember that max_new_tokens is a byte budget, not a token budget, so a limit that would produce a paragraph from a subword model produces only a sentence or two here.
Is your failure mode really tokenisation? Fine-tune the 1B on a few thousand examples of the task that trips your current model, whether character counting, transliteration, typo correction or code identifiers. Do the same with a subword model of similar size. If Bolmo 1B closes the gap, scaling up to the 7B, or to the byteified Bwen 8B, is a reasonable bet. If not, bytes are not your bottleneck.
No safety filtering. Check outputs. For larger options and instruction-tuned alternatives, browse the text generation directory, where Ai2's AstaBrief 8B is also listed.
Compare the model on the work you need to complete.
Bolmo 1B is listed for Text generation, Text-to-text AI models. It accepts text and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
The weights are downloadable under Apache-2.0. Check runtime support and size the complete model and working context for your expected concurrency.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Microsoft
Aion 1.0 Instruct 2.3B is Microsoft's MIT-licensed on-device small language model and Phi Silica successor for Windows, with a 32K context.
Gemini 4 Neon is Google's mid-tier Gemini 4 model for complex coding and analysis, sitting between Hydrogen and Carbon at $3/$15 per 1M tokens.
OpenAI
GPT-6.2 Astra is OpenAI's November 2026 flagship for hard professional, scientific and agentic work, at $8/$40 per million tokens with a 1.05M context.
OpenAI
GPT-6.2 Sol is OpenAI's everyday model for coding, agents and writing: $2/$10 per million tokens, 1.05M context, released 17 November 2026.
Alibaba
Qwen4 Max (qwen4-max) is Alibaba's first Qwen4-generation flagship: 1M-token context, text, image and video input, thinking on by default, $1.50/$4.50 per 1M.
Sarvam
Sarvam-105B-1711 is a refreshed open post-train of Sarvam's 105B MoE, tuned for tool calling, instruction following, agents and Indian languages.