impulse

Compare models

Up to 4 models side by side: list prices, what a month of real traffic costs, limits and capabilities. The best value in each row is marked with a check.

Clear
Model
Overview
Released26 Aug 20267 Oct 2026
Status
Older
Current
Open weightsYes, MITNo
Hosts
APIs you can call it on
19
Price · USD per million tokens, at the lab's own API where it sells the model
Priced atSarvamAnthropic
Input$0.144$0.10
Output$0.479$0.50
Cached input$0.0266$0.01
Batch
input / output
-$0.05 / $0.25
Long prompts
input / output
-$0.50 / $2.50
above 100K
Cheapest anywhere
input / output
$0.144 / $0.479
Sarvam
$0.10 / $0.50
Anthropic
Monthly bill · at the price above, 30 days, with each workload's cache and batch use
Chat assistant
5,000 a day, 2K in, 500 out
$68.46$59.40
RAG over documents
2,000 a day, 8K in, 600 out
$58.09$44.40
Coding agent
1,000 a day, 40K in, 2K out
$103$74.40
Overnight batch
20,000 a day, 4K in, 400 out
$460$180
Limits
Context window1M tokens1M tokens
Max output1M tokens128K tokens
Capabilities
Takes inText, ImagesText, Images, PDFs
ProducesTextText
Effort settings
as listed by the source
low, high, maxxhigh, max
Streaming
Function calling
JSON mode
Structured outputs
Reasoning
Tool calling
Prompt caching
Batch API
Adaptive thinking
Built-in web search
Computer use
Native structured output
Parallel tool calls