Granite-4.0-H-Small vs Llama 3.1 Nemotron Ultra 253B vs Apertus 8B
Too close to call on our weighted score (Llama 3.1 Nemotron Ultra 253B 65, Granite-4.0-H-Small 63, Apertus 8B 56). The right pick depends on what you value most.
IBMGranite-4.0-H-Small
63/100- ECI—
- Price$0.064 / $0.265
- Context131K
NVIDIALlama 3.1 Nemotron Ultra 253B
65/100- ECI—
- PriceFree / Free
- Context128K
Swiss AI
Apertus 8B
56/100- ECI—
- Price$0.10 / $0.20
- Context66K
Too close to call
It is close. Our weighted score puts them within 3 points (Llama 3.1 Nemotron Ultra 253B 65/100, Granite-4.0-H-Small 63/100, Apertus 8B 56/100), so choose by what matters most for your work: Llama 3.1 Nemotron Ultra 253B on price and Granite-4.0-H-Small for long inputs. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.
- CapabilityNot enough dataNo independent benchmark covers every model here yet
- Lowest priceLlama 3.1 Nemotron Ultra 253BLlama 3.1 Nemotron Ultra 253B Free · Granite-4.0-H-Small $0.114 · Apertus 8B $0.125 per 1M tokens (3:1 blend)
- Longest contextGranite-4.0-H-SmallGranite-4.0-H-Small 131,072 · Llama 3.1 Nemotron Ultra 253B 128,000 · Apertus 8B 65,536 tokens
- Widest inputsSame inputsGranite-4.0-H-Small: Text · Llama 3.1 Nemotron Ultra 253B: Text · Apertus 8B: Text
- Self-hostingAll open weightsEvery model here can be downloaded and run on your own hardware
| Measure | Weight | Granite-4.0-H-Small | Llama 3.1 Nemotron Ultra 253B | Apertus 8B |
|---|---|---|---|---|
| Price | 50% | 95 | 100 | 93 |
| Inputs & features | 30% | 35 | 35 | 25 |
| Context window | 20% | 24 | 24 | 12 |
| Overall | 100% | 63/100 | 65/100 | 56/100 |
Left out because at least one model lacks the data: capability. The remaining weights were rescaled.
Every spec in one table
Highlighted cells lead their row. Dashes mean the data is not published.
| Specification | Granite-4.0-H-SmallIBM | Llama 3.1 Nemotron Ultra 253BNVIDIA | Apertus 8BSwiss AI |
|---|---|---|---|
| Capability | |||
| Capabilities Index (ECI) | — | — | — |
| ECI rank | — | — | — |
| Price per million tokens | |||
| Input | $0.064 | Free (best) | $0.10 |
| Output | $0.265 | Free (best) | $0.20 |
| Cached input | — | — | — |
| Blended (3:1) | $0.114 | Free (best) | $0.125 |
| Long-context rate | Same rate | Same rate | Same rate |
| Price source | Official watsonx.ai API | Official Nvidia API | Median of 1 providers |
| Limits | |||
| Context window | 131,072 tokens (best) | 128,000 tokens | 65,536 tokens |
| Max output | 131,072 tokens (best) | 8,192 tokens | 8,192 tokens |
| Inputs and features | |||
| Text | Yes | Yes | Yes |
| Images | No | No | No |
| PDFs | No | No | No |
| Audio | No | No | No |
| Video | No | No | No |
| Reasoning | No | Yes | No |
| Tool calling | Yes | Yes | Yes |
| Structured output | Yes | No | No |
| Availability | |||
| Weights | Open | Open | OpenApache-2.0 |
| API model ID | ibm/granite-4-h-small | nvidia/llama-3.1-nemotron-ultra-253b-v1 | — |
| API providers | 1 | 1 | 1 |
| Released | Oct 2, 2025 | Apr 7, 2025 | Sep 2, 2025 |
| Knowledge cutoff | — | — | Sep 2025 |
What would a month cost?
Enter your expected volume in millions of tokens. List prices only; caching and batch discounts would lower these.
- Granite-4.0-H-Small$1.17
- Llama 3.1 Nemotron Ultra 253BFree
- Apertus 8B$1.40
Which should you choose?
Which is better: Granite-4.0-H-Small, Llama 3.1 Nemotron Ultra 253B or Apertus 8B?
It is close. Our weighted score puts them within 3 points (Llama 3.1 Nemotron Ultra 253B 65/100, Granite-4.0-H-Small 63/100, Apertus 8B 56/100), so choose by what matters most for your work: Llama 3.1 Nemotron Ultra 253B on price and Granite-4.0-H-Small for long inputs. The score weighs price 50%, inputs & features 30%, context window 20%. None of the shared benchmarks cover every model here yet, so capability is left out of this verdict.
Which is cheaper, Granite-4.0-H-Small, Llama 3.1 Nemotron Ultra 253B or Apertus 8B?
Llama 3.1 Nemotron Ultra 253B is cheaper at Free input / Free output per million tokens (official Nvidia API price). Granite-4.0-H-Small costs $0.064 input / $0.265 output per million tokens (official watsonx.ai API price); Apertus 8B costs $0.10 input / $0.20 output per million tokens (median across 1 API provider). Llama 3.1 Nemotron Ultra 253B is listed as free.
Which scores higher on benchmarks?
There is no independent benchmark that covers all three models yet. Granite-4.0-H-Small has not been scored yet, Llama 3.1 Nemotron Ultra 253B has not been scored yet and Apertus 8B has not been scored yet.
Which is better for coding?
There are no published SWE-bench Verified results for Granite-4.0-H-Small, Llama 3.1 Nemotron Ultra 253B and Apertus 8B yet, so there is no like-for-like coding score. Test both on a sample of your own tasks. All three support tool calling for agent workflows.
Which has the bigger context window?
Granite-4.0-H-Small has the largest context window at 131,072 tokens, against 128,000 for Llama 3.1 Nemotron Ultra 253B and 65,536 for Apertus 8B. Maximum output per response: Granite-4.0-H-Small up to 131,072, Llama 3.1 Nemotron Ultra 253B up to 8,192, Apertus 8B up to 8,192 tokens.
Which can read images, PDFs, audio or video?
Granite-4.0-H-Small accepts text; Llama 3.1 Nemotron Ultra 253B accepts text; Apertus 8B accepts text. They handle the same number of input types.
Are any of these open source?
Yes, all three publish their weights (Apache-2.0), so you can self-host them.
Which is newer?
Granite-4.0-H-Small is the newest, released Oct 2, 2025. Apertus 8B came out Sep 2, 2025; Llama 3.1 Nemotron Ultra 253B came out Apr 7, 2025. Knowledge cutoff: Apertus 8B Sep 2025.
How do you decide the winner?
Each model gets a 0–100 score on capability (50%, independent benchmark results); price (25%, blended price per million tokens (3 input : 1 output), log scale); inputs & features (15%, image, PDF, audio and video input, tool calling, structured output and reasoning); context window (10%, maximum tokens per request, log scale). Dimensions missing for any model are dropped and the remaining weights rescaled, so every model is judged on the same evidence. Specs and prices come from public model listings and the labs’ own API pages; capability scores come from independent benchmark runs. Data updated Oct 4, 2026.
