The Cheapest LLM for Translation in 2026 (Priced on a Real Job)

When optimizing large language model (LLM) costs, the nature of the task is paramount. While some tasks are input-heavy, others, like translation, are inherently output-heavy. This distinction fundamentally shifts which models offer the best value, as output token pricing often dominates the total cost.

For a real-world task of translating a 3,000-word document, our analysis reveals a staggering price disparity. The cheapest model completes this job for just $0.0002, while the most expensive costs $0.375. This represents a 2259x price spread, demonstrating that the cheapest LLM for general chat is often not the cheapest for translation when workload shape changes.

The Unique Cost Dynamics of Translation Tasks

Translation is characterized by generating an output that is roughly equivalent in length to the input. This means that a significant portion of the cost for such tasks comes from the generation of output tokens, rather than just processing input tokens. Models with competitive input pricing might become uneconomical if their output pricing is disproportionately high for such workloads.

Therefore, selecting the most cost-effective LLM for translation requires focusing on models that offer aggressive output token rates. This approach often leads to different top-ranking models compared to tasks like summarization or question answering, where input context might be larger than the desired output.

The Data: Cheapest LLMs for a 3,000-Word Translation

To identify the most cost-effective options, we priced a real job: translating a 3,000-word document. Out of 198 models analyzed, the pricing data provides a clear hierarchy based purely on cost efficiency for this specific task.

The cheapest model for this translation task is Ling-2.6-flash, coming in at an impressive $0.0002. This is a crucial finding for anyone looking to scale translation workflows without incurring prohibitive costs. The accompanying table provides a detailed breakdown of costs for various models, illustrating the wide range of pricing.

Task priced: Translate a ≈3,000-word document (output roughly matches input length). Representative sample of 198 models priced from QuoteFirst's live catalog (provider list prices, no markup). Get a quote for your own task at quotefirst.ai — prices update as providers change rates.

Understanding the Price Spread: From Cents to Fractions of Cents

The median cost for translating the 3,000-word document was $0.0058, represented by DeepSeek V3 0324. This still represents a significant saving compared to the most expensive model, Claude Opus 4.7, which would cost $0.375 for the same job. The 2259x price spread highlights the critical importance of model selection for output-heavy tasks.

This vast difference underscores that not all LLMs are priced equally for all tasks. Tools like QuoteFirst allow users to describe their specific task and receive real dollar quotes from a wide array of models, ensuring they identify the most economical solution tailored to their workload rather than relying on generalized pricing tiers.

Beyond Price: What About Quality and Performance?

While price is the primary focus for identifying the 'cheapest' LLM, performance and quality are also factors. The data shows some models, like Gemini 3 Flash, are noted for being "Extremely fast" and "Very low cost," or GPT-5.6 Luna for being "Very affordable" and offering "Fast responses." Even models with higher costs, such as Claude Opus 4.7, are noted for "Frontier reasoning and coding" and "Careful, nuanced writing," suggesting a trade-off in some cases.

However, for a straightforward translation task where output length is a direct match to input, the dominant factor for cost optimization remains the output token price. Users must balance their specific quality requirements against the significant cost savings offered by optimized models.

Frequently asked questions