impulse

Gemma4

Open weights
API id gemma4
Input / 1M
$0.372
at Sarvam
Output / 1M
$0.957
at Sarvam
Context window
262K
tokens
Max output
262K
tokens per reply
Hosts
1
1 listings
Released
-
date not published

Pricing

List prices at Sarvam, in US dollars per million tokens.

TierInput / 1MOutput / 1M
Standard$0.372$0.957
Cached input
71% off input
$0.106

What it costs a month

At Sarvam list prices, 30 days.

  • Chat assistant5,000 requests a day, 2K in / 0.5K out, cached$160
  • RAG over documents2,000 requests a day, 8K in / 0.6K out, cached$149
  • Coding agent1,000 requests a day, 40K in / 2K out, cached$281
  • Overnight batch20,000 requests a day, 4K in / 0.4K out$1,123
Use your own numbers

Where to run it

1 host serve Gemma4. The comparison is on blended price, 3 parts input to 1 part output.

HostIn / out per 1M
SarvamCloud
gemma4
$0.372 / $0.957

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
TextImages
Produces
Text
Features
Parallel tool callsPrompt caching
Open weights
License: Gemma License
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 Sarvam.

Smaller Google models

No cheaper Google model keeps its key capabilities.

Other frontier labs

  • GPT-5.4 Nano
    OpenAI · on OpenAI
    $0.20 / $1.25
    Has every key capability
    11% less
  • Claude Haiku 5.5
    Anthropic · on Anthropic
    $0.10 / $0.50
    Has every key capability
    61% less
  • GPT-6 Luna
    OpenAI · on OpenAI
    $0.10 / $0.50
    Has every key capability
    61% less
  • GPT-4o Mini
    OpenAI · on OpenAI
    $0.15 / $0.60
    Lacks reasoning
    49% less

Open-weight picks

  • Qwen3.7 Plus
    Alibaba Qwen · on Replicate
    $0.276 / $1.10
    Has every key capability
    7% less
  • MiMo V2.6 Pro
    Xiaomi · on GMI Cloud
    $0.348 / $0.696
    Has every key capability
    16% less
  • Muse Glimmer 30B
    Meta · on EmpirioLabs AI
    $0.20 / $0.80
    Has every key capability
    33% less
  • Deepseekv4.1 Flash
    DeepSeek · on Sarvam
    $0.149 / $0.585
    Has every key capability
    50% less

Call it

Using Sarvam's API.

Python
from litellm import completion

response = completion(
    model="sarvam/gemma4",
    messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)

Quick answers

How much does Gemma4 cost?

On Sarvam, Gemma4 costs $0.372 per million input tokens and $0.957 per million output tokens. Cached input costs $0.106 per million tokens.

Which APIs offer Gemma4?

Gemma4 runs on 1 API: Sarvam.

What is Gemma4's context window?

262K tokens, with up to 262K tokens in one reply.

Is Gemma4 being retired?

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

What are cheaper alternatives to Gemma4?

Models that cost less and keep most of what it can do: GPT-5.4 Nano on OpenAI (11% less), Claude Haiku 5.5 on Anthropic (61% less) and Qwen3.7 Plus on Replicate (7% less). Savings are on a mix of 3 input tokens to 1 output token.

What is Gemma4's API model id?

Use gemma4 with Sarvam.

Sources

Raw catalog entry for gemma4 on Sarvam
{
  "cache_read_input_token_cost": 1.063829787e-7,
  "input_cost_per_token": 3.723404255e-7,
  "litellm_provider": "sarvam",
  "max_input_tokens": 262144,
  "max_output_tokens": 262144,
  "mode": "chat",
  "output_cost_per_token": 9.574468085e-7,
  "source": "https://www.sarvam.ai/token-factory/pricing",
  "supports_audio_input": false,
  "supports_function_calling": true,
  "supports_parallel_function_calling": true,
  "supports_prompt_caching": true,
  "supports_reasoning": true,
  "supports_response_schema": true,
  "supports_system_messages": true,
  "supports_tool_choice": true,
  "supports_video_input": false,
  "supports_vision": true
}