Skip to main content
BenchLM
Data

LLM Price vs Performance Chart

Citable stat

As of November 12, 2026, Gemini 4 Carbon (Google) is the cheapest model within 90% of the overall leader's score: 89.2 points at $10.00 per million output tokens, 75% less than Claude Fable 5.2 (93.9 points, $40.00).

Pricing verified

Compare cost vs performance for ranked LLMs: each model’s active public score against its current API price. The shortlist favors the cheapest model that keeps at least 90% of the leader’s score; the efficiency frontier shows every undominated tradeoff.

  • 90% pick, overallGemini 4 CarbonGoogle89.2$10.00 per 1M output tokens75% below the leader’s price
  • Highest score, overallClaude Fable 5.2Anthropic93.9$40.00 per 1M output tokensAvailability restrictions apply
  • Price floor, overallQwen3.7 FlashAlibaba47.3$0.13 per 1M output tokensLowest price in this view

Best model for every budget

The highest overall score in each price band, with the next model in the band. Bands apply to the output price per million tokens.

The share compares each band’s top score with the highest score at any price.

Top overall scorer in each output price band
Budget per 1M output tokensTop scorer
Under $1Claude Haiku 5.567.7 points · $0.50 · 72% of the leaderNext: MiniMax M3.1 Flash (58.7)
$1 to $5Gemini 3.7 Flash65.9 points · $3.75 · 70% of the leaderNext: Muse Spark 1.2 (65.1)
$5 to $15Gemini 4 Carbon89.2 points · $10.00 · 95% of the leaderNext: Claude Sonnet 5.5 (83.1)
$15 and overClaude Fable 5.293.9 points · $40.00 · 100% of the leaderNext: Claude Opus 5.5 (86.6)

Score against price

Each point is a model, with price on the horizontal axis and score on the vertical, so the chart shows cost vs performance for every priced model in the view. The dark line is the efficiency frontier; for anything in the gray area under it, another model costs no more and scores at least as high. Prices use a log scale. Open a point for its details.

95 priced models match these filters. The blended view assumes three input tokens for every output token.

20406080100$0.10$0.50$1$5$10$50$100Price, $ per 1M tokens (log)LLM score against API priceOverall, output pricePricing as of November 12, 2026benchlm.ai
Efficiency frontierShaded: another model costs no more and scores at least as high90% pickOutline: estimated score

Select a point, or tab to it and press Enter, to see its price, its status and its nearest better value, the closest in price of the models that cost no more and score at least as high.

Best value by category

For each category with enough priced models, the leader, the cheapest model within 90% of its score, and what the pick saves. Output prices, one row per model family.

Select a category to show it on the chart.

Leader and 90% pick in each category, output prices
Category90% pick
OverallLeader: Claude Fable 5.2 (93.9, $40.00)Gemini 4 Carbon89.2 points · $10.00 · saves 75%
AgenticLeader: Claude Fable 5.2 (91.0, $40.00)Gemini 4 Carbon84.6 points · $10.00 · saves 75%
CodingLeader: Claude Fable 5.2 (89.8, $40.00)Gemini 4 Carbon84.0 points · $10.00 · saves 75%
ReasoningLeader: Claude Fable 5.2 (91.2, $40.00)Gemini 4 Carbon84.8 points · $10.00 · saves 75%
KnowledgeLeader: Claude Fable 5.2 (98.6, $40.00)Claude Opus 5.589.8 points · $20.00 · saves 50%
MultimodalLeader: Claude Fable 5.2 (90.6, $40.00)Claude Opus 5.585.1 points · $20.00 · saves 50%
Instruction FollowingLeader: Grok 4.3 (97.1, $2.50)MiniMax M393.8 points · $1.20 · saves 52%

All priced models

Every model on the chart, under the same filters, ranked by overall score. The status says whether another model costs no more and scores at least as high; the last column, nearest better value, names the closest such model in price.

95 models · 5 on the frontier

Every priced model in the current view with its score, prices and status
#ModelNearest better value
1Claude Fable 5.2Anthropic · Estimated93.9$40.00$16.00On the frontierNone
2Gemini 4 CarbonGoogle · Supported89.2$10.00$4.00On the frontierNone
3Claude Opus 5.5Anthropic · Supported86.6$20.00$8.00Within 90% of the leaderGemini 4 Carbon $10.00
4GPT-6 AstraOpenAI · Supported85.7$50.00$20.00Within 90% of the leaderClaude Fable 5.2 $40.00
5Claude Sonnet 5.5Anthropic · Supported83.1$10.00$4.001 model at the same price scores higherGemini 4 Carbon $10.00
6Claude Opus 5Anthropic · Supported76.1$25.00$10.003 cheaper models score at least as highClaude Opus 5.5 $20.00
7GPT-5.6 SolOpenAI · Supported72.5$20.00$8.003 models score at least as high for the same price or lessClaude Opus 5.5 $20.00
8Kimi K3.1 ProMoonshot AI · Estimated70.4$10.00$4.002 models at the same price score higherGemini 4 Carbon $10.00
9Kimi K3Moonshot AI · Supported68.5$15.00$6.003 cheaper models score at least as highGemini 4 Carbon $10.00
10Claude Haiku 5.5Anthropic · Supported67.7$0.50$0.20On the frontierNone
11Claude Opus 4.8Anthropic · Supported67.5$25.00$10.008 models score at least as high for the same price or lessClaude Opus 5 $25.00
12Muse WonderMeta · Estimated67.4$20.00$8.007 models score at least as high for the same price or lessClaude Opus 5.5 $20.00
13Grok 4.6xAI · Supported66.3$6.00$3.001 cheaper model scores at least as highClaude Haiku 5.5 $0.50
14GPT-5.5OpenAI · Supported66.0$30.00$11.2511 cheaper models score at least as highClaude Opus 5 $25.00
15Gemini 3.7 FlashGoogle · Supported65.9$3.75$1.501 cheaper model scores at least as highClaude Haiku 5.5 $0.50
16GPT-5.4OpenAI · Supported65.5$15.00$5.637 models score at least as high for the same price or lessKimi K3 $15.00
17Muse Spark 1.2Meta · Supported65.1$4.25$2.002 cheaper models score at least as highGemini 3.7 Flash $3.75
18DeepSeek V4.1 Flash 1017DeepSeek · Supported64.5$1.20$0.521 cheaper model scores at least as highClaude Haiku 5.5 $0.50
19Gemini 3.8 FlashGoogle · Supported64.5$3.75$1.503 models score at least as high for the same price or lessGemini 3.7 Flash $3.75
20Claude Sonnet 5Anthropic · Supported63.8$10.00$4.009 models score at least as high for the same price or lessGemini 4 Carbon $10.00

Prices per million tokens from the pricing catalog, verified November 12, 2026. Blended assumes three input tokens for every output token.

How to read this page

Two rules decide what the page calls good price to performance. Neither turns benchmark points into dollars.

Efficiency frontier
The models no other model beats on both price and score. For every model off the frontier, at least one other model costs no more and scores at least as high.
The 90% rule
The pick is the cheapest model that keeps at least 90% of the leader’s score. Score per dollar alone favors the cheapest weak models, because benchmark points are an index, not units of finished work.

Questions

Which LLM API is cheapest?

Among ranked models with a published API price, Qwen3.7 Flash (Alibaba) has the lowest output price: $0.13 per million output tokens, with an overall score of 47.3. On the 3:1 blended basis the cheapest is Qwen3.7 Flash at $0.06 per million tokens. The cheapest model is rarely the one to pick, because its score is often far below the leader's.

What is the cheapest LLM that is still good?

That depends on the score you need, so set a floor and read the budget table. Under $1 per million output tokens, the top overall score is Claude Haiku 5.5 at 67.7, 72% of the leader's. The 90% pick in the hero is one floor; the frontier rows in the full table show the others.

Which LLM has the best price to performance?

The 90% pick at the top of this page, if you want at least 90% of the overall leader’s score at the lowest price. For a different score floor, use the efficiency frontier: its 5 models in the overall view are the cost-efficient choices, since no other model beats any of them on both price and score.

What is the efficiency frontier?

The efficiency frontier is the set of models that no other model beats on both price and score. For a model below and to the right of it, another model costs no more and scores at least as high. In the overall view, 5 models are on it, from Qwen3.7 Flash at $0.13 to Claude Fable 5.2 at $40.00 per million output tokens.

Should I compare output price or blended price?

Output price is the default because output tokens cost more and dominate the bill for long answers and agent loops. The 3:1 blended price weighs three input tokens for every output token, which fits chat and retrieval work with long prompts. On blended price the overall 90% pick stays the same.

How are scores calculated?

The chart reads the active public ranking lane for each category and pairs that score with current catalog pricing. Overall, coding, and agentic views use the current BenchAlign methodology when enabled; other categories use the provisional composite. The 90% shortlist is a decision aid, not a claim that benchmark points convert directly into dollars.

Inside one provider the tradeoff is easiest to see: Gemini 3.8 Flash or Gemini 3.1 Pro is a price-for-capability decision within Google's own line.

Keep the value shortlist current

One weekly email when price changes or new benchmark evidence alter the models worth shortlisting.

Read a sample issue

Join 7,000+ subscribers.

One email each week. Unsubscribe anytime.