Guides → Playground & Guide → Cheapest Model - Best Value for Your Workload

Cheapest Model - Best Value for Your Workload

🚀 Open the calculator

Meet Marcus Lee. Senior Engineer told to 'use the cheap model' for a new feature. "Cheapest model is meaningless without context. Cheapest for WHAT?"

🔥 Switched to Haiku to save money. Quality dropped. Switched back. CFO unhappy.

The story

'Cheapest model' is the wrong question - 'cheapest model that hits my quality bar' is the right one. Gemini 3 Flash at $0.50/1M output is cheap. So is DeepSeek V3 at $0.27. Both 'fail' on certain tasks where Claude Haiku or GPT-5 Mini succeed. Cheapest only matters if quality clears the threshold.

Marcus's mistake: switched to the absolute cheapest tier without testing on his workload. Customer support classification - Haiku worked, saved 60%. But for the agentic workflow with tool calls, Haiku struggled with the schema and returned malformed JSON. Quality cost outweighed price savings.

Three buckets of 'cheap.' (1) Ultra-cheap (DeepSeek, Gemini Flash, Haiku) - fine for classification, simple Q&A, narrow extraction. (2) Mid-cheap (GPT-5 Mini, Sonnet 3.5) - solid for general agentic work, RAG, structured outputs. (3) Almost-frontier (Sonnet 4.6, GPT-5.5) - needed for complex reasoning, math, code generation. Pick the cheapest tier that passes your eval, not the cheapest model overall.

📊 CALCULATOR AT A GLANCE
🚀 Open the full calculator ✉️ Email [email protected]
Cheapest Model - Best Value for Your Workload full size

🎛 Inputs you control

Each input shapes the cost. Click an input on the calculator to set it — explanations below match the live calculator field by field.

▸ What are you building — Determines minimum-quality bar AND the typical token shape (input/output sizes) the recommendation is costed against.
How to choose: Pick the closest match to YOUR primary use case. If you have multiple use cases, run the calc once per use case — they often select different optimal models.
▸ Tier-1 provider only — When checked, limits to OpenAI / Anthropic / Google. Excludes DeepSeek, xAI, Mistral, and others that may not have your required compliance posture.
How to choose: Check this only if you have an actual constraint (data residency rule, BAA requirement, procurement allowlist). Otherwise leave unchecked — Tier-1 models are typically 2-10× more expensive than the absolute cheapest.
▸ Vision required — Filters to models that can process images.
How to choose: Only check if vision is required for your workload. Vision-capable variants of mid-tier models often cost more than the text-only variant.
▸ Agent-capable required — Excludes Nano-tier and Flash-Lite models that lack the reasoning depth for multi-turn tool use.
How to choose: Check this for workloads with iterative tool calls, planning, or multi-step reasoning. Skip for one-shot tasks (classification, simple extraction, single-turn chat).

📊 Outputs computed for you

What you'll see after the calculator runs. Each card explains how to read the number.

▸ Cheapest-first model ranking — Models meeting ALL your constraints, ordered cheapest to most expensive at the use case's typical token shape.
How to read: Top entry is the cheapest valid option. Don't default to it blindly — read the "why this works" reasons. If the cheapest is a brand-new or unfamiliar provider, the 2nd-cheapest from a familiar vendor is often the pragmatic choice.
▸ Per-model rationales — Workload-specific reasons each model survives your filters: caching applicability, batch eligibility, long-context handling, free-tier availability, compliance posture.
How to read: A "Prompt caching 90% off" note means effective price is dramatically lower than sticker for repeat-context workloads. A "Batch API (50% off)" note means async-eligible traffic gets half-price.
▸ Per-call cost at typical shape — Dollar cost per single API call using the use case's default input/output token shape. With cache/batch savings applied where relevant.
How to read: Multiply by your expected daily request count for daily total. If your actual token shape differs from the preset, recompute in Cost Calculator with your real numbers.

About this calculator: Cheapest Model - Best Value for Your Workload

Cheapest LLM for your workload - by tier, by task type, by quality threshold. Updated daily as vendors shift pricing. Beyond the per-token table.

Inputs you control

Input Impact on result Range Typical
Required quality (1-10) How critical is quality? 1 = noise OK. 5 = decent. 7 = production. 9 = mission-critical. 10 = no failure tolerance. 1 – 10 6
Tasks per day Volume - at 50K+/day, the per-task gap × volume becomes meaningful. 10 – 10M 50000
Task complexity (1-10) 1 = classification. 3 = simple Q&A. 5 = RAG over docs. 7 = multi-step reasoning. 10 = research-grade analysis. 1 – 10 5

Outputs computed for you · model: token

Output How inputs affect it
Monthly cost ($) computed from inputs
Annual cost ($) monthlyUsd × 12

Below: live sliders. Move them to see numbers change in real time. * Output uses the generic compute model — for precise numbers use the full calculator below.

What you're looking at

Each input shapes your cost. Move the slider — see the impact.

6

How critical is quality? 1 = noise OK. 5 = decent. 7 = production. 9 = mission-critical. 10 = no failure tolerance.

Estimated: —
50,000

Volume - at 50K+/day, the per-task gap × volume becomes meaningful.

Estimated: —
5

1 = classification. 3 = simple Q&A. 5 = RAG over docs. 7 = multi-step reasoning. 10 = research-grade analysis.

Estimated: —

Ready to run the numbers?

Open the full calculator — pick a model, enter your tokens, see per-call, daily, monthly, and annual cost.

🚀 Open the full calculator →

Reading your result

Match complexity score to model tier. Score 1-3: ultra-cheap fine. 4-6: mid-cheap. 7-9: balanced/premium. 10: frontier only.

Quality threshold acts as a floor. Required 7+ quality eliminates ultra-cheap tier regardless of complexity. Required 9+ may eliminate mid-cheap too.

Volume amplifies savings - and risks. At 50K/day, picking 30% cheaper-per-task = $X/mo saved. Picking 5% lower-quality = customer complaints. Test before scaling.

What "good" looks like:
  • Ultra-cheap winners now: Gemini 3 Flash, DeepSeek V3, Haiku 4.5 - for tasks scoring complexity ≤4 and quality ≤6
  • Mid-cheap winners: GPT-5 Mini, Claude Haiku - for complexity 4-6, quality 6-8
  • Best value at premium quality: Sonnet 4.6 (often beats Opus on cost-quality ratio)
  • Avoid: Frontier-only for tasks scoring complexity ≤6 - wastes 5-10× the cost

Cheapest LLMs right now (per 1M tokens)

Verified 20 hours ago
  1. 1
    Command R7b 12.2024
    $0.150 in · $0.037 out ·
  2. 2
    voxtral-mini
    $0.040 in · $0.040 out ·
  3. 3
    ministral-3-3b
    $0.100 in · $0.100 out ·

Three real scenarios

Same calculator, three different team sizes. Click a tab to see how the numbers shift.

$5,302 / month ≈ $63,618 / year

Sentiment classification on user reviews. Low complexity, low quality bar. Ultra-cheap tier wins by 80% over mid-cheap. ~$300/mo at 100K/day.

Healthy range: DeepSeek V3 / Gemini Flash, ~$300/mo

See inputs used
qualityThreshold
5
tasksPerDay
100,000
complexityScore
2
inputTokensPerTask
800
outputTokensPerTask
50

Trade-offs

Cost isn't the only dimension. Click any constraint — see how recommendations change.

What matters most to you? Click any dimension — recommendations update.

Best fit for "cost":

  1. Ultra-cheap (DeepSeek, Gemini Flash) $0.10-0.50/1M output
  2. Mid-cheap (Haiku 4.5, GPT-5 Mini) $1-3/1M output
  3. Balanced (Sonnet 4.6) $15/1M - best value at quality

Cost ranks change weekly. Anchor on tier, not vendor. Re-check pricing quarterly because rankings shift as vendors compete.

Use cases

Pre-loaded scenarios for the most common applications. Click a tab to see realistic numbers — then the "Try this scenario" button to load it into the calculator above.

$35,692 / month ≈ $428,305 / year

High-volume tool selection (which API to call?). Mid-cheap with prompt caching cuts cost dramatically. Tool defs cache → 80% input discount.

Healthy range: Haiku + caching, ~$2.5K/mo

See inputs used
qualityThreshold
7
tasksPerDay
200,000
complexityScore
5
inputTokensPerTask
3,000
outputTokensPerTask
100

What this calculator can't tell you

Honest limitations — every model is wrong; some are useful. Where this one falls short:

For these, use: Multi-Model Router for routing strategy. Cost Calculator for full bill.

Where to go next

Route per-query for stacked savings →

Use cheap for easy, premium for hard.

 Project full bill with chosen model →

Validate the savings.

 FT a cheap model for narrow tasks →

Cheap + FT often beats premium + prompting.

Methodology

Source
/ai-cost-economics
Extraction
Per-vendor pricing pulled daily. Quality benchmarks from LMSYS Arena, MMLU, HumanEval.
Editorial gate
8-layer defense — see aicost.ai/ai-cost-economics
Last verified
6/4/2026, 8:00:00 PM

Author: Subu Vdaygiri, Founder & CEO of CloudIntelligence.ai. 17 years Fortune 100 (Ingram Micro, Siemens). Wharton CTO program · Kellogg CPO program · 10× AWS+Azure certified.

3 years of pricing history

Why this matters: pricing for major vendors has dropped 40-90% in the last 24 months. A budget set 12 months ago is probably wrong by 30%+.

View 3-year history for →
📖 Data sources & methodology 161 text models · 9 embeddings · 24 vision · 41 audio · 8 vector DBs across 10 vendor pages · last verified 2026-06-05

Methodology

  • All prices are USD per 1 million tokens, current as of 2026-06-05.
  • Vendor-published values have no mark. Inferred/extrapolated values are marked with * and listed below.
  • Batch API discounts are 50% off standard rates across providers that offer Batch mode.
  • Prompt caching discounts vary by provider (typically 80-90% off cached input tokens).
  • Regional data-residency surcharges (Anthropic 1.1x, OpenAI 1.1x, Google regional tiers) are NOT included in base rates.
  • Long-context pricing tiers apply when input exceeds model threshold.
  • Embedding prices are input-only (no output tokens generated).

Primary sources

Last-verified date is the most recent successful daily snapshot (aicost_pricing_snapshots) or, when no snapshot exists yet, the latest successful crawler run (aicost_crawler_runs). 10 of 10 vendors are currently verified. Aggregator services (TokenCost, AI Pricing Guru, etc.) are not listed.

Anthropic
2026-06-05
https://www.anthropic.com/pricing
Daily snapshot since Sep 2023 · 578 days captured
Anthropic Docs
2026-06-05
https://platform.claude.com/docs/en/about-claude/pricing
Daily snapshot since Sep 2023 · 578 days captured
OpenAI
2026-06-05
https://openai.com/api/pricing/
Daily snapshot since Sep 2023 · 579 days captured
Google AI
2026-06-05
https://ai.google.dev/gemini-api/docs/pricing
Daily snapshot since Dec 2023 · 554 days captured
Google Vertex
2026-06-05
https://cloud.google.com/vertex-ai/generative-ai/pricing
Daily snapshot since Dec 2023 · 554 days captured
DeepSeek
2026-06-05
https://api-docs.deepseek.com/quick_start/pricing
Daily snapshot since May 2024 · 493 days captured
xAI
2026-06-05
https://x.ai/api
Daily snapshot since Nov 2024 · 411 days captured
Mistral
2026-06-05
https://mistral.ai/pricing
Daily snapshot since Dec 2023 · 552 days captured
Cohere
2026-06-05
https://cohere.com/pricing
Daily snapshot since Sep 2023 · 578 days captured
Voyage AI
2026-06-05
https://docs.voyageai.com/docs/pricing

Inferred values (marked with * in calculator tables)

Derived from industry conventions, not directly published by the vendor. Typical conventions: cached input = 10% of base (90% off), Batch API = 50% of base (50% off).

Vendor / Model Field Why it’s inferred
Anthropic — Claude Sonnet 4.6 cachedInput Derived at 10% of input rate — Anthropic publishes 90% cache-hit discount on this tier.
Anthropic — Claude Sonnet 4.5 cachedInput Derived at 10% of input rate; same 90% cache-hit convention as Sonnet 4.6.
Anthropic — Claude Sonnet 4.5 batchInput Derived at 50% of standard input — Anthropic documents uniform 50% Batch discount.
Anthropic — Claude Sonnet 4.5 batchOutput Derived at 50% of standard output — Anthropic documents uniform 50% Batch discount.
Anthropic — Claude Haiku 4.5 cachedInput Derived at 10% of input rate — Anthropic 90% cache-hit discount convention.
OpenAI — GPT-5.4 Mini cachedInput Derived at 10% of input — OpenAI documents automatic 90% discount on cache hits across GPT-5.x tier.
OpenAI — GPT-5.4 Nano cachedInput Derived at 10% of input — OpenAI 90% cache-hit convention.
OpenAI — GPT-5.4 Nano batchInput Derived at 50% of input — OpenAI Batch API uniform 50% discount.
OpenAI — GPT-5.4 Nano batchOutput Derived at 50% of output — OpenAI Batch API uniform 50% discount.
OpenAI — GPT-5.4 Pro cachedInput Derived at 10% of input — OpenAI 90% cache-hit convention.
OpenAI — GPT-5.4 Pro batchInput Derived at 50% of input — OpenAI Batch API uniform 50% discount.
OpenAI — GPT-5.4 Pro batchOutput Derived at 50% of output — OpenAI Batch API uniform 50% discount.
OpenAI — GPT-5.2 cachedInput Derived at 10% of input; no residency uplift.
OpenAI — GPT-5.2 batchInput Derived at 50% of input.
OpenAI — GPT-5.2 batchOutput Derived at 50% of output.
OpenAI — GPT-5 cachedInput Derived at 10% of input.
OpenAI — GPT-5 batchInput Derived at 50% of input.
OpenAI — GPT-5 batchOutput Derived at 50% of output.
OpenAI — GPT-5.5 Pro cachedInput Derived at 10% of input — OpenAI does not publish a cached rate for *-pro models; using the family convention.
OpenAI — GPT-5.5 Pro batchInput Derived at 50% of input.
OpenAI — GPT-5.5 Pro batchOutput Derived at 50% of output.
OpenAI — GPT-5.2 Pro cachedInput Derived at 10% of input — pro-tier convention.
OpenAI — GPT-5.2 Pro batchInput Derived at 50% of input.
OpenAI — GPT-5.2 Pro batchOutput Derived at 50% of output.
OpenAI — GPT-5.1 batchInput Derived at 50% of input.
OpenAI — GPT-5.1 batchOutput Derived at 50% of output.
OpenAI — GPT-5 Pro batchInput Derived at 50% of input.
OpenAI — GPT-5 Pro batchOutput Derived at 50% of output.
OpenAI — GPT-5 Nano cachedInput Derived at 10% of input.
OpenAI — GPT-5 Nano batchInput Derived at 50% of input.
OpenAI — GPT-5 Nano batchOutput Derived at 50% of output.
Google — Gemini 3 Flash cachedInput Derived at 10% of input — Google caching discount convention ~90%.
Google — Gemini 3.1 Flash-Lite cachedInput Derived at 10% of input — Google caching convention.
Google — Gemini 3.1 Flash-Lite batchInput Derived at 50% of input — Google Batch API uniform 50% discount.
Google — Gemini 3.1 Flash-Lite batchOutput Derived at 50% of output — Google Batch API uniform 50% discount.
Google — Gemini 2.5 Pro cachedInput Derived at 10% of input.
Google — Gemini 2.5 Flash cachedInput Derived at 10% of input.
Google — Gemini 2.5 Flash-Lite cachedInput Derived at 10% of input — Google caching convention.
Google — Gemini 2.5 Flash-Lite batchInput Derived at 50% of input — Google Batch API uniform 50% discount.
Google — Gemini 2.5 Flash-Lite batchOutput Derived at 50% of output — Google Batch API uniform 50% discount.
Google — Gemini 2.0 Flash cachedInput Derived at 25% of input per Google 2.0 family caching rates.
Google — Gemini 2.0 Flash batchInput Derived at 50% of input — Google Batch API uniform 50% discount.
Google — Gemini 2.0 Flash batchOutput Derived at 50% of output — Google Batch API uniform 50% discount.
Google — Gemini 2.0 Flash-Lite cachedInput Derived at 10% of input — Google caching convention.
Google — Gemini 2.0 Flash-Lite batchInput Derived at 50% of input — Google Batch API uniform 50% discount.
Google — Gemini 2.0 Flash-Lite batchOutput Derived at 50% of output — Google Batch API uniform 50% discount.
xAI — Grok 4 (legacy) cachedInput Extrapolated at 25% of base.

Pricing is cross-verified against the LiteLLM community registry when available. Daily snapshots are kept in aicost_pricing_snapshots; every change is logged to aicost_price_changelog with old & new values for full audit trail. Read the full methodology →