Frontier · ≥ 50
≈ GPT-5.5 / Opus-class
100.00
- Champion
- Gemini-3.5 Flash
- Blended
- $3.3750/1M
- Capability
- 50
- Qualifiers
- 5
Indices
IFX answers “are list prices themselves falling?” with a conservative held-basket Laspeyres. CoI answers “is it getting cheaper to buy a given level of capability?” with a substitution measure. The two indices are complementary.
Base 100 on 2024-04-01 · reconstructed (dashed) + live (solid)
IFX · per-era waterfall
Five eras chained price-relative. Each row shows the era's start and end levels and the chained move.
base 100 · 2024-04-01
| Era | Window | Start | End | Move |
|---|---|---|---|---|
A Modern frontier GPT-4 Turbo, Claude 3 family, Gemini 1.5; the first era where the modern frontier vs budget tier structure became legible. | 2024-04-30 → 2024-09-30 | 100.00 | 86.16 | -13.84% |
B Reasoning emerges o1, Claude 3.5 Sonnet (new), DeepSeek V3 + R1 establish reasoning as a distinct price tier. | 2024-10-31 → 2025-03-31 | 86.16 | 86.16 | 0.00% |
C1 GPT-5 era opening Claude Opus 4 / Sonnet 4, Gemini 2.5 Pro / Flash, o3, GPT-4.1. The o3 list-price cut on Jun 10 2025 (−80%) is the single biggest sourced move in the reconstruction. | 2025-04-30 → 2025-09-30 | 86.16 | 80.69 | -6.35% |
C2 GPT-5 era closing GPT-5 family, Claude Opus 4.5 / Sonnet 4.5 / Haiku 4.5, Gemini 3 family, broad open-weight MoE generation. Matched-basket prices held flat. | 2025-10-31 → 2026-03-31 | 80.69 | 80.69 | 0.00% |
D Current The live 29-model basket. Daily collection began 2026-06-18. Sub-indices for input-only and output-only token prices are live-only from this point forward. | 2026-04-30 → 2026-06-21 | 80.69 | 80.69 | 0.00% |
Documented sourced events
Reconstruction is matched-sample — only documented price changes on held basket members move the level. List-price stickiness is what the index reveals.
2024-08-13 Era A
−13.84% on the chained index
Google halves Gemini 1.5 Pro and Flash
Gemini 1.5 Pro: $3.50/$10.50 → $1.25/$5.00 (blended −58%). Gemini 1.5 Flash: $0.35/$1.05 → $0.075/$0.30 (blended −79%). Combined with GPT-4o's Aug 6 cost-down (−42% blended), this was the largest IFX move of 2024.
2025-06-10 Era C1
−6.35% on the chained index
OpenAI cuts o3 list price 80%
o3 input $10 → $2, output $40 → $8 — an 80% list-price cut across the board. At 7% basket weight in C1's frontier tier, this single event drove the largest sourced chained-index move in the reconstruction.
Cost of Intelligence · bar detail
Each bar tracks the cheapest blended $/1M-token model whose canonical AA Intelligence Index score meets the gate. The cheaper the champion, the lower the bar's base-100 level.
scale aa-intelligence-index-v4.1 · base 100 on 2026-06-18
Frontier · ≥ 50
100.00
Capable · ≥ 40
100.00
Budget · ≥ 25
100.00
Label is approximate reference class for public display; the gate is the numeric threshold. The CoI scale is pinned to AA Intelligence Index v4.1 — when AA bumps the scale, a chain-splice is recorded rather than the bar being silently moved.
IFX-Video · sibling index
USD per generated second at 720p/24fps — basket of 8 models. Base 100 on 2026-06-21.
frontier · cap ≥ 80
Sora 2
OpenAI
$0.100/s
5 qualifiers
workhorse · cap ≥ 70
Hailuo 02
MiniMax
$0.045/s
10 qualifiers
budget · cap ≥ 50
LTX 2.3
Lightricks
$0.020/s
15 qualifiers
| Model | Provider | Tier | Capability | $/sec | Weight |
|---|---|---|---|---|---|
| Veo 3.1 Lite | budget | 60 | $0.030/s | 10.00% | |
| Vidu Q2 | Shengshu | budget | 54 | $0.037/s | 10.00% |
| Hailuo 02 | MiniMax | workhorse | 74 | $0.045/s | 13.33% |
| Sora 2 | OpenAI | frontier | 84 | $0.100/s | 13.33% |
| Kling 3.0 Turbo | Kuaishou | frontier | 86 | $0.120/s | 13.34% |
| Runway Gen-4 | Runway | workhorse | 78 | $0.120/s | 13.33% |
| Luma Ray3.14 | Luma Labs | workhorse | 76 | $0.200/s | 13.34% |
| Veo 3.1 | frontier | 92 | $0.400/s | 13.33% | |
| Universe (tracked, not in basket) | |||||
| HunyuanVideo 1.5 | Tencent | universe | 56 | $0/s (self-hosted) | — |
| LTX 2.3 | Lightricks | universe | 60 | $0.020/s | — |
| Wan 2.5 | Alibaba | universe | 68 | $0.050/s | — |
| Pika 2.2 | Pika Labs | universe | 60 | $0.080/s | — |
| Seedance 2.0 | ByteDance | universe | 75 | $0.124/s | — |
| Kling 2.6 | Kuaishou | universe | 84 | $0.160/s | — |
| Moonvalley Marey | Moonvalley | universe | 70 | $0.180/s | — |
| Sora 2 Pro | OpenAI | universe | 88 | $0.500/s | — |
IFX-Video tracks the cost of generated video, priced in USD per generated second at a 720p / 24fps reference. The basket weighting mirrors the LLM IFX (frontier 40% / workhorse 40% / budget 20%, equal within tiers). Where a provider sells credits, subscription seats, or per-clip charges (Runway, Kling, Pika, Luma), the listed $/sec is the blended effective rate at 720p on the cheapest seat that exposes the model. Per-bar champions are the cheapest models that clear an AA Video Arena capability gate (frontier ≥80, workhorse ≥70, budget ≥50). The index anchors at base 100 on the sibling inception date — the LLM IFX's 2024-04 inception does NOT apply because the underlying unit ($/sec) is incommensurable with $/1M tokens. Refresh cadence is weekly in v1 (video pricing moves slower than LLM pricing); the contract still publishes a daily entry.
IFX-Image · sibling index
USD per generated image at 1024×1024 standard — basket of 8 models. Base 100 on 2026-06-21.
frontier · cap ≥ 80
HiDream-O1-Image (Dev)
HiDream AI
$0.030/img
5 qualifiers
workhorse · cap ≥ 70
Luma Photon 1
Luma Labs
$0.015/img
13 qualifiers
budget · cap ≥ 50
Luma Photon Flash
Luma Labs
$0.0020/img
19 qualifiers
| Model | Provider | Tier | Capability | $/img | Weight |
|---|---|---|---|---|---|
| Luma Photon Flash | Luma Labs | budget | 54 | $0.0020/img | 10.00% |
| Nano Banana (Gemini 2.5 Flash Image) | budget | 68 | $0.039/img | 10.00% | |
| GPT Image 2 | OpenAI | frontier | 86 | $0.040/img | 13.33% |
| FLUX.2 [pro] | Black Forest Labs | workhorse | 78 | $0.040/img | 13.33% |
| Imagen 4 | Google DeepMind | workhorse | 76 | $0.040/img | 13.33% |
| Seedream 5.0 | ByteDance | frontier | 88 | $0.050/img | 13.34% |
| Ideogram 3.0 | Ideogram | workhorse | 72 | $0.060/img | 13.34% |
| Nano Banana Pro (Gemini 3 Pro Image) | frontier | 92 | $0.130/img | 13.33% | |
| Universe (tracked, not in basket) | |||||
| Midjourney V8.1 | Midjourney | universe | 84 | $0/img (sub-only) | — |
| Qwen-Image 2.0 | Alibaba | universe | 58 | $0.0050/img | — |
| FLUX.2 [klein] | Black Forest Labs | universe | 60 | $0.010/img | — |
| GPT Image 1 Mini | OpenAI | universe | 54 | $0.011/img | — |
| Stable Diffusion 3.5 Large | Stability AI | universe | 52 | $0.012/img | — |
| Luma Photon 1 | Luma Labs | universe | 70 | $0.015/img | — |
| FLUX.2 [flex] | Black Forest Labs | universe | 74 | $0.020/img | — |
| Reve 2.0 | Reve AI | universe | 71 | $0.020/img | — |
| HiDream-O1-Image (Dev) | HiDream AI | universe | 82 | $0.030/img | — |
| Recraft V4 | Recraft | universe | 72 | $0.040/img | — |
| Adobe Firefly Image Model 4 | Adobe | universe | 70 | $0.060/img | — |
| Imagen 4 Ultra | Google DeepMind | universe | 84 | $0.080/img | — |
IFX-Image tracks the cost of image generation, priced in USD per generated 1024×1024 image at standard quality. Basket weighting mirrors IFX-Video (frontier 40% / workhorse 40% / budget 20%, equal within tiers). For models with multiple quality tiers (GPT Image 2 spans $0.005–$0.211 etc.), the listed $/image is the mid/standard tier; the range is documented in each basket entry's note. Subscription-only services (Midjourney) are deliberately excluded from the basket because effective $/image depends entirely on usage volume — they're kept in the universe for editorial reference. Per-bar champions are the cheapest models that clear an AA Image Arena capability gate (frontier ≥80, workhorse ≥70, budget ≥50). The index anchors at base 100 on the sibling inception date.