impulse

Granite 4.2 8B

Open weights

Granite 4.2 8B is a dense reasoning model from IBM. It is suited for mathematics, code generation, multilingual dialogue, and agentic workflows that need multi-step reasoning.

API id granite-4.2-8bReleased 31 Aug 2026 (Aug 2026)
Input / 1M
$0.06
at W&B Inference
Output / 1M
$0.15
at W&B Inference
Context window
131K
tokens
Max output
-
tokens per reply
Hosts
2
2 listings
Released
Aug 2026
31 Aug 2026

Pricing

List prices at W&B Inference, in US dollars per million tokens.

TierInput / 1MOutput / 1M
Standard$0.06$0.15
Cached input
75% off input
$0.015

What it costs a month

At W&B Inference list prices, 30 days.

  • Chat assistant5,000 requests a day, 2K in / 0.5K out, cached$25.20
  • RAG over documents2,000 requests a day, 8K in / 0.6K out, cached$23.40
  • Coding agent1,000 requests a day, 40K in / 2K out, cached$43.20
  • Overnight batch20,000 requests a day, 4K in / 0.4K out$180
Use your own numbers

Where to run it

2 hosts serve Granite 4.2 8B. The comparison is on blended price, 3 parts input to 1 part output.

HostIn / out per 1M
W&B InferenceInference
granite-4.2-8b
$0.06 / $0.15
ReplicateInference
granite-4.2-8b
$0.06 / $0.25
+30% vs headline

Capabilities

Core

  • Streaming
  • Function calling
  • JSON mode
  • Structured outputs

Multimodal

  • Vision
  • Image input
  • Audio
  • Image generation
  • Video

Advanced

  • Reasoning
  • Tool calling
  • MCP
  • Prompt caching
  • Batch API
  • Embeddings
Takes in
Text
Produces
Text
Features
Prompt caching
Open weights
License: Apache-2.0
Runs in about 21.1 GB of GPU memory
Runtimes: Ollama, llama.cpp, vLLM

Cheaper alternatives

Current models that cost less and keep most of what this one can do: tools, vision, reasoning, structured output, caching and enough context.

Same model, cheaper host

No host sells it for less than W&B Inference.

Smaller IBM models

No cheaper IBM model keeps its key capabilities.

Other frontier labs

No cheaper match from another frontier lab.

Open-weight picks

  • DeepSeek V4.1 Flash Fast
    DeepSeek · on Refuseless
    $0.05 / $0.15
    Has every key capability
    9% less
  • Qwen3.7 Flash Variant1
    Alibaba Qwen · on EmpirioLabs AI
    $0.028 / $0.11
    Has every key capability
    41% less
  • Qwen3.5 4B
    Alibaba Qwen · on EmpirioLabs AI
    $0.04 / $0.07
    Has every key capability
    42% less

Model line

Newer and older versions, at each lab's own price.

  1. Aug 2026Granite 4.2 8BLatest$0.06 / $0.15
  2. UndatedGranite 4.1 8B$0.05 / $0.10

Call it

Using W&B Inference's API.

Python
from litellm import completion

response = completion(
    model="wandb/granite-4.2-8b",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

Quick answers

How much does Granite 4.2 8B cost?

On W&B Inference, Granite 4.2 8B costs $0.06 per million input tokens and $0.15 per million output tokens. Cached input costs $0.015 per million tokens.

Which APIs offer Granite 4.2 8B?

Granite 4.2 8B runs on 2 APIs: W&B Inference and Replicate.

What is Granite 4.2 8B's context window?

131K tokens.

Is Granite 4.2 8B being retired?

No retirement date is scheduled for it on any API we track.

Is there a newer version of Granite 4.2 8B?

No. Granite 4.2 8B is the newest model in its line.

What are cheaper alternatives to Granite 4.2 8B?

Models that cost less and keep most of what it can do: DeepSeek V4.1 Flash Fast on Refuseless (9% less). Savings are on a mix of 3 input tokens to 1 output token.

When was Granite 4.2 8B released?

IBM released it on 31 Aug 2026.

What is Granite 4.2 8B's API model id?

Use granite-4.2-8b with W&B Inference.

Sources

Raw catalog entry for granite-4.2-8b on W&B Inference
{
  "litellm_provider": "wandb",
  "mode": "chat",
  "supports_reasoning": true,
  "input_cost_per_token": 6e-8,
  "output_cost_per_token": 1.5e-7,
  "cache_read_input_token_cost": 1.5e-8,
  "max_input_tokens": 131000,
  "supports_prompt_caching": true,
  "source": "https://www.coreweave.com/products/coreweave-forge/serverless-inference"
}