The economics of intelligence
Which AI model gives you the most brain for the money?
66 models in our dated snapshot, charted by how smart they are against what they cost: per token, per task, and what the same work would cost billed by usage.
Capability snapshot: 2026-11-01. 6 of 66 models have task measurements. OpenAI task measurements are pending.
Compare a subscription with usage-based pricing
Same work, billed by usage
Not measured yet
What would the same work cost billed by usage? We have not verified a usage sample for this tier yet.
Its price alone does not tell us how much work it includes.
Measurement target: Codex Desktop only. Current capacity is not measured; this does not measure ChatGPT web chat.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Plan price checked 2026-11-04. public USD offer; account availability unverified How we compare.
Same work, billed by usage
Not measured yet
What would the same work cost billed by usage? We have not verified a usage sample for this tier yet.
Its price alone does not tell us how much work it includes.
Measurement target: Codex Desktop only. Current capacity is not measured; this does not measure ChatGPT web chat.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Plan price checked 2026-11-04. public USD offer; account availability unverified How we compare.
Same work, billed by usage
Not measured yet
What would the same work cost billed by usage? We have not verified a usage sample for this tier yet.
Its price alone does not tell us how much work it includes.
Measurement target: Codex Desktop only. Current capacity is not measured; this does not measure ChatGPT web chat.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Plan price checked 2026-11-04. public USD offer; account availability unverified How we compare.
Same work, billed by usage
Not measured yet
What would the same work cost billed by usage? We have not verified a usage sample for this tier yet.
Its price alone does not tell us how much work it includes.
Measurement target: Codex Desktop only. Current capacity is not measured; this does not measure ChatGPT web chat.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Plan price checked 2026-11-04. public USD offer; account availability unverified How we compare.
Same work, billed by usage
Not measured yet
What would the same work cost billed by usage? We have not verified a usage sample for this tier yet.
Its price alone does not tell us how much work it includes.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Plan price checked 2026-11-04. public USD offer; account availability unverified How we compare.
Same work, billed by usage
Not measured yet
What would the same work cost billed by usage? We have not verified a usage sample for this tier yet.
Its price alone does not tell us how much work it includes.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Plan price checked 2026-11-04. public USD offer; account availability unverified How we compare.
Same work, billed by usage
Not measured yet
What would the same work cost billed by usage? We have not verified a usage sample for this tier yet.
Its price alone does not tell us how much work it includes.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Plan price checked 2026-11-04. public USD offer; account availability unverified How we compare.
Same work, billed by usage
Not measured yet
What would the same work cost billed by usage? We have not verified a usage sample for this tier yet.
Its price alone does not tell us how much work it includes.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Plan price checked 2026-11-04. public USD offer; account availability unverified How we compare.
Same work, billed by usage
Not measured yet
What would the same work cost billed by usage? We have not verified a usage sample for this tier yet.
Its price alone does not tell us how much work it includes.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Plan price checked 2026-11-04. public USD offer; account availability unverified How we compare.
Same work, billed by usage
Not measured yet
What would the same work cost billed by usage? We have not verified a usage sample for this tier yet.
Its price alone does not tell us how much work it includes.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Plan price checked 2026-11-04. public USD offer; account availability unverified How we compare.
Same work, billed by usage
Not measured yet
What would the same work cost billed by usage? We have not verified a usage sample for this tier yet.
Its price alone does not tell us how much work it includes.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Plan price checked 2026-11-04. public USD offer; account availability unverified How we compare.
Same work, billed by usage
Not measured yet
What would the same work cost billed by usage? We have not verified a usage sample for this tier yet.
Its price alone does not tell us how much work it includes.
PUBLIC SOURCE / NOT TESTED
Gemini Apps for personal accounts, web and mobile: standard baseline; Plus 2× standard; Pro 4× standard; Ultra 5× Pro or 20× Pro, depending on the offer.
Google describes a five-hour refresh with a weekly cap. These are its relative compute units, not prompt, token or API-dollar allowances. No Google subscription was tested.
The Ultra limit variant attached to this catalogue price is unverified. Prices stay in the separate verified catalogue; API-dollar capacity and task shares stay unavailable.
Google's published limits, checked 2026-11-04. Scope and model labels.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Plan price checked 2026-11-04. US public page How we compare.
Same work, billed by usage
Not measured yet
What would the same work cost billed by usage? We have not verified a usage sample for this tier yet.
Its price alone does not tell us how much work it includes.
PUBLIC SOURCE / NOT TESTED
Gemini Apps for personal accounts, web and mobile: standard baseline; Plus 2× standard; Pro 4× standard; Ultra 5× Pro or 20× Pro, depending on the offer.
Google describes a five-hour refresh with a weekly cap. These are its relative compute units, not prompt, token or API-dollar allowances. No Google subscription was tested.
The Ultra limit variant attached to this catalogue price is unverified. Prices stay in the separate verified catalogue; API-dollar capacity and task shares stay unavailable.
Google's published limits, checked 2026-11-04. Scope and model labels.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Plan price checked 2026-11-04. US public page How we compare.
Historical what-if · October 20, 2026 UTC
Not a current allowance.
Projected API-equivalent per week
≈ $2,577.44
Recorded work scaled to a full weekly meter; not measured full-week capacity.
$2,061.95–$3,436.59 · meter rounding only.
How this estimate works
20 Oct 2026 UTC sample, repriced at 4 Nov standard short-context API rates. Scaled from logged usage and a rounded meter, assuming all meter movement belongs to the logged work and the same workload remains repeatable. Outside usage and meter weighting are unverified. Not measured full-week capacity or current-plan value. Range covers rounding only.
Mixed Opus/Haiku and cache-heavy work; recovered known gap usage, but browser/cloud and other-project usage may be missing. Historical Max 20x only.
Oct 20, 2026, 01:27 UTC to Oct 20, 2026, 02:28 UTC.
Logged work priced at approximately $103.10 using the 4 November tariff. Meter: 3 → 7 (4 percentage points).
The $200.00 monthly fee is the historical test-plan assumption, not a verified invoice.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Fee shown is the historical test-plan price. API tariffs checked 2026-11-04. How we compare.
Historical what-if · October 20, 2026 UTC
Not a current allowance.
Projected API-equivalent per week
≈ $59.75
Recorded work scaled to a full weekly meter; not measured full-week capacity.
$55.48–$64.73 · meter rounding only.
How this estimate works
20 Oct 2026 UTC sample, repriced at 4 Nov standard short-context API rates. Scaled from logged usage and a rounded meter, assuming all meter movement belongs to the logged work and the same workload remains repeatable. Outside usage and meter weighting are unverified. Not measured full-week capacity or current-plan value. Range covers rounding only.
Codex CLI coding work, not ChatGPT web chat or Desktop. Recorded Astra/Luna/Sol work only; usage between recorded runs is not fully accounted for. Historical Plus price assumption only; no Pro multiplier.
Oct 20, 2026, 04:51 UTC to Oct 20, 2026, 05:22 UTC.
Logged work priced at approximately $7.77 using the 4 November tariff. Meter: 22 → 35 (13 percentage points).
The $20.00 monthly fee is the historical test-plan assumption, not a verified invoice.
API-equivalent use is the cost of the same work at usage-based prices, not cash credits or a guaranteed allowance.
Fee shown is the historical test-plan price. API tariffs checked 2026-11-04. How we compare.
All 66 models
Cheapest model scoring 160 or more: Claude Haiku 5.5, at $0.20 per million tokens.
Showing all 66 models. Scroll the table sideways to see every column.
| Model | Lab | ECI | Price / M tokens | Cost / task |
|---|---|---|---|---|
| GPT-6 Astra | OpenAI | 169.1 | $20 | not measured |
| Claude Opus 5.5 | Anthropic | 168.4 | $8 | $0.20 |
| GPT-6.1 Terra | OpenAI | 168.1 | $2 | not measured |
| GPT-6.1 Sol | OpenAI | 167.8 | $4 | not measured |
| Claude Sonnet 5.5 | Anthropic | 166.7 | $4 | $0.10 |
| Claude Fable 5.1 | Anthropic | 166.4 | $20 | $0.52 |
| Claude Opus 5 | Anthropic | 164.5 | $10 | $0.34 |
| GPT-6 Sol | OpenAI | 163.8 | $4 | not measured |
| GPT-5.5 Pro | OpenAI | 163.1 | $67.50 | not measured |
| Claude Fable 5 | Anthropic | 163.0 | $20 | $0.66 |
| GPT-5.6 Sol | OpenAI | 162.6 | $8 | not measured |
| Muse Wonder | Meta AI | 161.7 | $8 | not measured |
| Claude Haiku 5.5 | Anthropic | 160.4 | $0.20 | $0.0078 |
| GPT-5.6 Terra | OpenAI | 160.1 | $4.50 | not measured |
| GPT-5.4 Pro | OpenAI | 159.8 | $67.50 | not measured |
| Claude Opus 4.8 | Anthropic | 159.6 | $10 | not measured |
| GPT-5.5 | OpenAI | 159.4 | $11.25 | not measured |
| DeepSeek V4.1 Flash 1017 | DeepSeek | 158.8 | $0.53 | not measured |
| Kimi K3 | Moonshot | 158.3 | $6 | not measured |
| Gemini 3.7 Flash | Google DeepMind | 158.2 | $1.50 | not measured |
| Qwen 3.8 Max | Alibaba | 158.0 | $3 | not measured |
| Gemini 3.8 Flash | Google DeepMind | 157.6 | $1.50 | not measured |
| Muse Spark 1.3 | Meta AI | 157.4 | $2 | not measured |
| GPT-5.4 | OpenAI | 157.1 | $5.63 | not measured |
| Claude Sonnet 5 | Anthropic | 157.0 | $4 | not measured |
| Grok 4.6 | xAI | 156.9 | $3 | withheld |
| GPT-6 Luna | OpenAI | 156.8 | $0.20 | not measured |
| Claude Opus 4.7 | Anthropic | 156.7 | $10 | not measured |
| GPT-5.6 Luna | OpenAI | 156.7 | $0.45 | not measured |
| Qwen3.8 Max (0902) | Alibaba | 156.2 | $3 | not measured |
| GLM-5.3 | Z.ai (Zhipu AI) | 156.2 | $2.15 | not measured |
| DeepSeek V4 Pro 0813 | DeepSeek | 155.9 | $1.98 | not measured |
| Claude Opus 4.6 | Anthropic | 155.8 | $10 | not measured |
| Gemini 3.1 Pro | Google DeepMind | 155.2 | $4.50 | not measured |
| Muse Spark 1.2 | Meta AI | 155.2 | $2 | not measured |
| Gemini 3.5 Flash | Google DeepMind | 155.0 | $3.38 | not measured |
| Gemini 3.6 Flash | Google DeepMind | 154.7 | $1.50 | not measured |
| Grok 4.7 | xAI | 154.6 | $3 | withheld |
| Grok 4.5 | xAI | 154.4 | $3 | withheld |
| Qwen3.7-Max | Alibaba | 154.2 | $3.75 | not measured |
| Muse Spark 1.1 | Meta AI | 154.2 | $2 | not measured |
| GLM-5.3-Flash | Z.ai (Zhipu AI) | 153.2 | $0.24 | not measured |
| Claude Sonnet 4.6 | Anthropic | 152.9 | $6 | not measured |
| Grok 4.20 | xAI | 152.3 | $1.56 | not measured |
| GLM-5.2 | Z.ai (Zhipu AI) | 152.1 | $2.15 | not measured |
| Qwen 3.8 27B | Alibaba | 151.8 | $1.13 | not measured |
| Kimi K2.6 | Moonshot | 151.1 | $1.71 | not measured |
| Mistral Large 4 | Mistral AI | 150.6 | $1.05 | not measured |
| Inkling-Small | Thinking Machines | 150.5 | $0.53 | not measured |
| Kimi K2.7 Code | Moonshot | 150.2 | $1.71 | not measured |
| GLM-5.1 | Z.ai (Zhipu AI) | 149.9 | $2.15 | not measured |
| GPT-5.4 Mini | OpenAI | 148.8 | $1.69 | not measured |
| Inkling | Thinking Machines | 148.5 | $1.76 | not measured |
| Qwen 3.6 Plus | Alibaba | 147.8 | $1.13 | not measured |
| Qwen3.7-Plus | Alibaba | 147.1 | $0.70 | not measured |
| MiniMax-M3 | MiniMax | 146.9 | $0.53 | not measured |
| Qwen3.5 397B-A17B | Alibaba | 146.8 | $1.35 | not measured |
| Qwen3.6 27B | Alibaba | 146.3 | $1.35 | not measured |
| GPT-5.4 Nano | OpenAI | 146.1 | $0.46 | not measured |
| GLM-5 | Z.ai (Zhipu AI) | 145.8 | $1.55 | not measured |
| Gemini 3.5 Flash-Lite | Google DeepMind | 145.0 | $0.85 | not measured |
| Gemini 3.1 Flash-Lite | 144.3 | $0.56 | not measured | |
| Qwen3.7 Flash | Alibaba | 144.1 | $0.06 | not measured |
| Qwen 3.6 35B-A3B | Alibaba | 143.8 | $0.84 | not measured |
| Qwen3.5-35B-A3B | Alibaba | 142.6 | $0.69 | not measured |
| Gemma 4 26B A4B | Google DeepMind | 141.9 | $0.26 | not measured |
| No models match these filters. Clear filters to see all models. | ||||