ModelPriceWatch.com
Last scan 2026-09-29 Models tracked 272 Providers 35 Cheapest paid Granite 4.0 H Micro $0.017/Mtok in Every price links to its source

The LLM Price Index

One number for the cost of frontier intelligence — the blended price of a million tokens across today's flagship models, re-verified daily.

The LLM Price Index · as of

$5.34/1M tok

blended price of a million tokens across 10 frontier flagships — one current flagship per major lab.

16.8% since flat 25 days 3:1 in:out blend equal-weight 10 constituents

The index level has been unchanged for 25 days — it last moved Sep 4, 2026, and that move was a basket rebalance, not a reprice. A flat level is the finding, not a missed update: frontier list prices are structurally sticky and step only at a rebalance.

MODELPRICEWATCH.COM · 2026-09-29

The level over time

chain-linked · rescaled so today's point equals the headline

Index level — · 98 readings

2026-07-18: GPT-5.5 → GPT-5.6 Sol, Grok 4 → Grok 4.5 2026-07-25: Claude Opus 4.8 → Claude Opus 5, Qwen3-Max → Qwen3.7-Max, Moonshot added to the basket (Kimi K3) 2026-08-01: Meta added to the basket (Muse Spark 1.1) 2026-08-04: Qwen3.7-Max → Qwen3.8-Max 2026-08-06: Muse Spark 1.1 → Muse Spark 1.2 2026-08-12: Grok 4.5 → Grok 4.6 2026-08-18: GLM-5.2 → GLM-5.3 2026-09-04: GPT-5.6 Sol → GPT-6 Astra 2026-09-05: Muse Spark 1.2 → Muse Spark 1.3
index level · $/1M blended basket rebalance
$4.57 · 16.8% across 98 readings $5.34 ·
Basket rebalance events: the date each flagship slot changed occupant or opened, which model succeeded which, and whether the index level stepped.
Date Slot succession Effect
GPT-5.5 → GPT-5.6 Sol; Grok 4 → Grok 4.5 rebalance
Claude Opus 4.8 → Claude Opus 5; Qwen3-Max → Qwen3.7-Max; + Moonshot slot opened (Kimi K3) rebalance
+ Meta slot opened (Muse Spark 1.1) composition
Qwen3.7-Max → Qwen3.8-Max rebalance
Muse Spark 1.1 → Muse Spark 1.2 like-for-like
Grok 4.5 → Grok 4.6 like-for-like
GLM-5.2 → GLM-5.3 like-for-like
GPT-5.6 Sol → GPT-6 Astra rebalance
Muse Spark 1.2 → Muse Spark 1.3 like-for-like
The level steps only when a lab rebalances the basket to a differently-priced flagship — a rebalance. A like-for-like succession swaps in a lab's newer flagship at the same list price, so the level holds. A composition change adds a lab that previously had no slot: the chain-linked trend is unaffected by design (no phantom cliff), but the headline mean is now taken over a larger basket — a change in what is measured, not a market move.

The Intelligence Cost Curve

ceiling vs floor · log scale · GPQA Diamond (raw %)

The flagship index is the ceiling — what the frontier costs, and it barely moves. The floor is the cheapest model that still clears a fixed capability bar (GPQA Diamond ≥ 70, “GPT-4-class general reasoning”, anchored to GPT-4.1 (GPQA Diamond 70.0)). As cheaper capable models launch, the floor falls — the widening gap is the deflation story.

Ceiling vs floor —

47.3× cheaper

GPT‑4-class intelligence now costs $0.113/Mtok (DeepSeek V4 Flash) against the $5.34 flagship ceiling.

ceiling — flagship index, flat floor — cheapest GPT-4-class general reasoning model, falling
floor $1.93 · ceiling $5.34 · 10-flagship basket floor $0.113 ·

What stepped the floor down

o4-mini
$1.93/Mtok
Grok 4.3
$1.56/Mtok
DeepSeek V4 Pro
$0.544/Mtok
GPT-OSS 120B
$0.262/Mtok
DeepSeek V4 Flash
$0.113/Mtok
Floor uses a single absolute metric (GPQA Diamond (raw %), absolute 0–100), external only (self_reported excluded), deduped to base model (min blended across hosted SKUs), blended 3:1. Series begins — the first date a qualifying model had both a verified score and a price; earlier benchmark dates reconstructed from recorded sources are labelled as such. Full methodology → MODELPRICEWATCH.COM · 2026-09-29

What moved

latest refresh ·

No frontier flagship list-price moves since the last refresh — the basket held steady. That's the norm: frontier list prices are structurally sticky and step only at rebalances.

Cut three ways

sub-indices — each tells a different story

Cheapest frontier token

$0.750 /1M blended

Mistral Large 3 — how little frontier-class capability can cost.

Cost per intelligence point

$0.066

Blended cost ÷ benchmark score, over the 10/10 constituents with a published score. Best value: Mistral Large 3.

Open vs closed spread

$5.65 vs $4.08

Closed-weights vs open-weights flagships — the premium for proprietary models.

What's in the index

fixed basket — changes only on a deliberate rebalance
The LLM Price Index basket: each constituent model with provider, blended price per million tokens, and weights openness.
Model Provider Blended* $/1M Weights
GPT-6 Astra OpenAI $20.00 Closed
Claude Opus 5 Anthropic $10.00 Closed
Gemini 3.1 Pro Google $4.50 Closed
Grok 4.6 xAI $3.00 Closed
DeepSeek V4 Pro DeepSeek $1.98 Closed
Qwen3.8-Max Alibaba $3.00 Closed
Mistral Large 3 Mistral $0.750 Closed
GLM-5.3 Z.AI $2.15 Open
Kimi K3 Moonshot $6.00 Open
Muse Spark 1.3 Meta $2.00 Closed
* Blended $/1M = (3×input + output) ÷ 4. Methodology MODELPRICEWATCH.COM · 2026-09-29

Cite it

"Frontier intelligence costs $5.34 per million tokens — 16.8% since Feb 23, 2026, flat for 25 days."

The LLM Price Index — ModelPriceWatch, 2026-09-29 · modelpricewatch.com/price-index/

Citing a past date? Levels are restated onto the current basket when composition changes — the as-published archive keeps every dated headline frozen on its own basis. What we promise →

State of LLM Pricing — September 2026 →

The dated report: what the index, the capability floor, and the frontier spread showed on September 1 — and the two reprices that ended the sticky ceiling.

How it's calculated →

Full methodology: constituents, weighting, the 3:1 blend, rebalancing.

Get price alerts →

Email me when a model's price drops — free, no spam.

Use the data →

Free, CORS-open API. Cite the LLM Price Index freely.