Frontier LLM prices sat flat all October. Then two cheaper flagships took seats, and one doubles on December 1

The LLM Price Index didn't move once in October. Same ten flagships, same ten list prices, from October 1 to November 1. That's the first full calendar month without a move since the index's daily record began in August, and by November 1 it had been flat for 32 days, the longest stretch on that record.

The market wasn't flat. Two labs launched new flagships priced below the models holding their seats, and on November 7 both took those seats. The index fell 4.9%, to $4.89 per million tokens.

One of those two models is cheap only until November 30.

October in one table

The LLM Price Index from ModelPriceWatch is the equal-weight average of ten flagships' blended price per million tokens (3 parts input to 1 part output, list prices as printed on each vendor's own pricing page), one model per lab.

Sort the October 1 and November 1 baskets by price and all ten numbers are identical: $20.00 at the top (GPT-6 Astra), $0.75 at the bottom (Mistral Large 3), $3.00 median.

Why the cheaper flagships took five weeks to count

Gemini 4 Argon went on sale October 13 and Kimi K3.1 Pro on October 23, both at $4.00 blended. Google calls Argon "our flagship reasoning model"; Moonshot lists K3.1 Pro first in its K3 series.

Each index seat follows a lab's flagship line by name (Gemini Pro for Google, Kimi K for Moonshot), and both models launched under new names. A model with a new name enters only through a deliberate, dated rebalance, the same way GPT-6 Astra took OpenAI's seat in September. That rebalance happened on November 7.

The arithmetic on November 1 prices:

The December 1 catch

Gemini 4 Argon's $2 input / $10 output is an introductory price that runs through November 30. From December 1, Google's rate card prints $4/$20, or $8.00 blended. That's 78% above Gemini 3.1 Pro, the model it replaced in the index.

All else equal, that one repricing adds $0.40 to the index on December 1: $4.89 → $5.29. That would leave it 2.9% above the $5.14 it held all October, so November's 4.9% fall is already scheduled to reverse.

Argon isn't the only launch price with an end date:

• Gemini 4 Carbon, announced October 28, uses the same pattern: $2/$10 introductory, $4/$20 to follow. It's limited to a subset of cyber defenders and trusted partners and isn't on any Google rate card, so we don't list it.
• Gemini 3.6 Flash and 3.8 Flash keep their introductory $0.75/$3.75 through December 31 and are scheduled to double on January 1, 2027.
• Mistral Large 4 (preview) showed three prices in its first month: $0.68/$2.09 at launch on October 6 (shown beside a struck-through $1.36/$4.18 on October 12 and 19), $1.36/$4.18 alone on October 26, and $0.70/$2.10 from October 29.

If you're benchmarking a new model on cost, check whether the price you're testing at has an end date.

Every GA successor launched cheaper or flat

October releases from the index's labs that followed a model already on sale:

*Prompts up to 100K tokens. Longer prompts bill at $0.50/$2.50 ($1.00 blended), still half of Haiku 4.5.
†Moonshot also sells Kimi K3.1 Pro HighSpeed, the same model served faster, at $8.00 blended, 33% above Kimi K3.

Claude Haiku 5.5 is the month's headline: $0.10/$0.50, a tenth of Haiku 4.5's $1/$5, and Haiku 4.5 is still on sale at that price. Meta's Muse Wonder (preview, $4/$20) is a new tier above its Muse Spark line rather than a successor, at four times Muse Spark 1.3.

For the second month running, labs cut prices by launching new models, not by repricing old ones. If you're still on the previous model in a line, the newer one is probably cheaper.

Retirements are price changes too

OpenAI shut down eight of the models we track on October 23: GPT-3.5 Turbo, GPT-4 (0613), GPT-4 Turbo, GPT-4.1 nano, o1, o1-pro, o3-mini and o4-mini.

A shutdown doesn't change a list price, but if you're pinned to a retiring model you pay whatever its replacement costs. OpenAI's suggested replacements cut the bill for its old large models and raise it for its small ones:

*GPT-5.6 Sol's $4/$20 is a promotion OpenAI says runs "at least through November 21, 2026", with no end date. Its earlier $5/$30 is $11.25 blended, still cheaper than all three. o1-pro users are pointed to GPT-5.6 Sol in pro reasoning mode.

The suggested replacement isn't the only option. GPT-6.1 Terra ($2.00) launched in October at less than half of GPT-5.6 Terra.

October cleared out elsewhere too: Alibaba retired five tracked models including Qwen3-Max (October 10), Google deprecated Gemini 3.5 Flash and 3.7 Flash and routes their calls to 3.6/3.8 Flash (October 8), and Mistral deprecated Large 3 and Medium 3.5 for retirement on April 30, 2027.

The calendar to put in your notes

Two numbers that didn't move

Cache reads at 5% of input are now common at the top. Gemini 4 Argon ($0.10 vs $2.00) and GPT-6.1 Terra ($0.05 vs $1.00) launched there, joining Claude Opus 5.5 and GPT-6.1 Sol from September. Claude Haiku 5.5 and Kimi K3.1 Pro launched at 10%. If your prompts repeat, this matters more than anything in a blended-price index.

GPT-4-class capability still cost $0.090/Mtok. The index's companion floor, the cheapest model with an externally measured GPQA Diamond score of 70 or higher (roughly GPT-4-class reasoning), didn't move in October: DeepSeek V4 Flash on DeepInfra, 57.1x below the flagship ceiling. None of October's launches had an external GPQA Diamond score by November 1, so none could challenge it.

What to take from this

• A flat index can hide a moving market. October's 32 days of no change covered two flagship launches cheaper than the models they'd replace.
• Check the end date on a launch price. Argon, Carbon and the Gemini Flash models all launched with introductory rates. Budget for the rate card price, not the intro.
• Check your replacement before the shutdown date. For OpenAI's small retired models, the suggested replacement costs 2.3x to 6x more. A newer, cheaper model may fit.
• Moving to the newest model in a line usually saves money. Haiku 4.5 → 5.5 is a 90% cut for prompts up to 100K tokens.

Figures for October are as of November 1, 2026, and are fixed in the dated report at modelpricewatch.com/reports/state-of-llm-pricing-2026-11; the November 7 rebalance is logged in the methodology. The live index is at modelpricewatch.com/price-index, and the full price history dataset is on Hugging Face (CC-BY-4.0). Every price links to the vendor page it came from, with a capture timestamp.

Disclosure: published by ModelPriceWatch.

Did OpenAI's October shutdowns move you onto a pricier model, or did you find something cheaper than the replacement OpenAI suggested?