Open index · multi-turn · prompt caching · July 2026
The LLM Conversation Cost Index
A standard 10-turn conversation — 2,000-token system prompt, 1,000-token tool block, 200 user + 500 assistant tokens per turn, with the full history resent every turn — costs from $8.35 (GPT-4.1 nano) to $885 (Claude Fable 5) per 1,000 conversations as of July 2026. Prompt caching cuts that total 48–60% (input cost 64–88%).
Every LLM price page prices one prompt. This is the open index for the number nobody owns: what a real, multi-turn conversation costs once you account for the history being resent on every turn — priced per 1,000 conversations, per model, with and without caching. It is the reference table behind our Conversation Cost Simulator. Download the full dataset (CSV + JSON, CC-BY) below.
GPT-4.1 nano
Claude Fable 5
Claude Sonnet 5
prompt caching
The index
25-second explainer — why a real conversation costs far more than one prompt.
Cost per 1,000 standard conversations
The standard conversation is a fixed, reproducible unit (defined in Methodology): 10 turns, 63,500 total input tokens, 5,000 output tokens. Costs use each provider's published July 2026 list price. Click a column header to sort.
| Model | Provider | Input $/1M |
Output $/1M |
Cost / conversation |
Cost / 1,000 conv |
With caching / 1,000 conv |
Caching savings |
|---|---|---|---|---|---|---|---|
| GPT-4.1 nano | OpenAI | $0.10 | $0.40 | $0.0083 | $8.35 | $4.30 | 48.5% |
| Gemini 3 Flash | $0.50 | $3.00 | $0.0467 | $46.75 | $18.89 | 59.6% | |
| Claude Haiku 4.5 | Anthropic | $1.00 | $5.00 | $0.0885 | $88.50 | $42.27 | 52.2% |
| GPT-5.6 Luna | OpenAI | $1.00 | $6.00 | $0.0935 | $93.50 | $47.27 | 49.4% |
| Gemini 3.5 Flash | $1.50 | $9.00 | $0.1403 | $140.25 | $56.66 | 59.6% | |
| Claude Sonnet 5 | Anthropic | $2.00 | $10.00 | $0.1770 | $177.00 | $84.55 | 52.2% |
| Gemini 3.1 Pro | $2.00 | $12.00 | $0.1870 | $187.00 | $75.55 | 59.6% | |
| GPT-5.6 Terra | OpenAI | $2.50 | $15.00 | $0.2338 | $233.75 | $118.19 | 49.4% |
| Claude Opus 4.8 | Anthropic | $5.00 | $25.00 | $0.4425 | $442.50 | $211.37 | 52.2% |
| GPT-5.6 Sol | OpenAI | $5.00 | $30.00 | $0.4675 | $467.50 | $236.38 | 49.4% |
| Claude Fable 5 | Anthropic | $10.00 | $50.00 | $0.8850 | $885.00 | $422.75 | 52.2% |
All figures USD. "With caching" applies each provider's cache-read / cache-write rates to the stable prefix; output is never cached. Snapshot of published list prices, verified 2026-07-13.
Open data
Download the dataset
Free to use, cite and redistribute under CC-BY 4.0 — just link back. Both files carry the same rows shown above plus the raw token totals and caching multipliers, so you can reproduce or extend the index.
11 rows · spreadsheet-ready JSON llm-conversation-cost-index.json
rows + methodology + metadata
Attribution: Data: The LLM Conversation Cost Index by waynetools.rdkworld.com, CC-BY — https://waynetools.rdkworld.com/llm-conversation-cost-index/
Embed this badge (free)
Drop this live cost badge on your blog, docs or dashboard. It always shows the current number and links back to the open data — no script, no tracking.
<iframe src="https://waynetools.rdkworld.com/embed/cost-badge.html" width="360" height="200" style="border:0" title="LLM Conversation Cost — waynetools" loading="lazy"></iframe>Methodology
How the index is calculated
An LLM API is stateless: to continue a conversation the client resends the entire history as input every turn. So the input tokens on turn t are system + tools + (t−1)·(user + assistant) + user. The standard conversation fixes every variable so all models are compared identically:
- System prompt: 2,000 tokens (resent every turn)
- Tools / context block: 1,000 tokens (resent every turn)
- User message: 200 tokens per turn
- Assistant response: 500 tokens per turn
- Turns: 10
That yields 63,500 input tokens and 5,000 output tokens per conversation. Cost per conversation prices those at each model's list rate; cost per 1,000 multiplies by 1,000. With caching, turn 1 writes the whole prefix and each later turn reads the prior prefix (at the provider's cache-read rate) and writes only the new user+assistant tokens (at the cache-write rate) — 54,000 cache-read and 9,500 cache-write token-equivalents across the conversation. Output is never cached, on any provider.
Without caching Input, turn 1 = 2,000 + 1,000 + 200 = 3,200 tok Input, turn 10 = 2,000 + 1,000 + 9x700 + 200 = 9,500 tok Total input over 10 turns = 63,500 tok -> $0.3175 (@ $5 / 1M) Total output over 10 turns = 5,000 tok -> $0.1250 (@ $25 / 1M) Cost per conversation = $0.4425 -> $442.50 / 1,000 With prompt caching Cache reads = 54,000 tok x 0.1 = 5,400 tok-equiv Cache writes = 9,500 tok x 1.25 = 11,875 tok-equiv Input (cached) = 17,275 x $5 / 1M = $0.0864 (input down 73%) Output (never cached) = $0.1250 Cost per conversation = $0.2114 -> $211.37 / 1,000 Saved by caching = 52% of total cost
Price sources (verified July 13, 2026)
List prices are a dated snapshot taken from each provider's official pricing page. Providers change prices without notice — always confirm against the source:
- OpenAI (GPT-5.6 Sol / Terra / Luna, GPT-4.1 nano): openai.com/api/pricing
- Anthropic (Claude Fable 5, Opus 4.8, Sonnet 5, Haiku 4.5): anthropic.com/pricing · prompt caching docs
- Google (Gemini 3.1 Pro, 3.5 Flash, 3 Flash): ai.google.dev/gemini-api/docs/pricing
Caveats. This is a planning index, not a bill. It assumes fixed average token sizes, a cache hit on every turn (turns within the cache TTL), standard synchronous pricing (no batch discounts, no context-length premiums, and no Gemini per-hour cache storage fee). Real conversations vary; verify against your provider's usage dashboard.
Get a price-change alert
OpenAI, Anthropic and Google change prices without warning. Get one email the day any major model's price moves — plus a fresh cost cheat sheet each month. Free, no spam.
One list, unsubscribe anytime. We never share your address.
How to read it
Three ways teams use this index
Size a monthly bill
Multiply the per-1,000 figure by your expected conversation volume. 50k conversations/month on Sonnet 5 ≈ $8,850 — or ~$4,200 with caching wired up.
Pick a tier
The value frontier is Sonnet 5 and Gemini 3.1 Pro near $180/1,000. Premium reasoning (Opus, GPT-5.6 Sol, Fable 5) costs 2.5–5× that for the same conversation shape.
Value caching
Caching claws back roughly half the total bill on every model here. On long or high-traffic conversations it is almost always worth the wiring.
FAQ
Common questions
How much does a multi-turn LLM conversation cost?+
A standard 10-turn conversation (2,000-token system prompt, 1,000-token tool block, 200 user and 500 assistant tokens per turn, with the full history resent every turn) costs from about $8.35 per 1,000 conversations on GPT-4.1 nano to about $885 per 1,000 on Claude Fable 5, as of July 2026 list prices. Mid-tier workhorses land around $88–$233 per 1,000. Prompt caching cuts these totals by roughly 48–60%.
Which model is cheapest for chatbots and multi-turn agents?+
On the standard conversation, GPT-4.1 nano is cheapest (~$8.35/1,000), then Gemini 3 Flash (~$46.75) and Claude Haiku 4.5 (~$88.50). Among frontier-quality models, Claude Sonnet 5 (~$177) and Gemini 3.1 Pro (~$187) are the value tier; GPT-5.6 Sol, Claude Opus 4.8 and Claude Fable 5 sit above $440 per 1,000.
How much does prompt caching save on a conversation?+
About 48–60% of total cost on the standard conversation, depending on the provider. Because caching only discounts the repeated input prefix and never the generated output, input-cost savings are larger — roughly 64–88% — but output tokens dilute the total. Gemini's cheap cache writes give it the largest total-cost cut here.
Why does a 10-turn conversation cost so much more than one prompt?+
Because LLM APIs are stateless: to continue a conversation the client resends the entire history as input on every turn. The standard conversation bills 63,500 input tokens over 10 turns even though a single turn is only ~3,200 — the same system prompt, tool block and early messages are paid for again and again. Cumulative input cost grows with roughly the square of the turn count.
What is the standard conversation used in this index?+
A fixed, reproducible unit: a 2,000-token system prompt and 1,000-token tool/context block (both resent every turn), plus 200 user and 500 assistant tokens per turn, over 10 turns — 63,500 input tokens and 5,000 output tokens total. Cost per conversation prices those at each model's list rate; cost per 1,000 multiplies by 1,000. See Methodology.
Is this the exact bill I will pay?+
No — it is a planning index on a fixed synthetic conversation and published list prices, so every model is compared on the same footing. Real conversations vary in message length, tool calls and thinking tokens (billed as output on reasoning models), and caching savings depend on breakpoint placement and traffic timing. Use it to rank models and size a budget, then confirm against your provider's usage dashboard.
More waynetools