RXed AI News

AI to the bone.

The graphs

The pricing table answers “what does this model cost”; these charts answer “what is happening to prices”. Rebuilt from the same data every night. Every chart has a table view, and every caveat is printed, not hidden.

Data updated 2026-08-17. Capability & cost data: Artificial Analysis

Price vs capability

The same reasoning score sells at prices an order of magnitude apart — the labelled models are the cheapest at their capability level.

frontierheavymediumlight
204060$0.10$0.30$1.00$3.00$10.0AA reasoning index →$/M blended (log)Claude Opus 5Grok 4.6Muse Spark 1.2Gemini 3.7 FlashGPT-5.6 LunaHunyuan Hy3Ling 3.0 Flash

Capability = Artificial Analysis Intelligence Index; price = $/M tokens blended 3:1 input:output on list prices. Darker = heavier weight class. Log price axis.

Table view
ModelClassReasoning$/M blended
Claude Opus 5frontier63.1$10.0
Claude Fable 5frontier62.1$20.0
GPT-5.6 Solfrontier60.9$11.2
Grok 4.6frontier60.9$3.00
Kimi K3frontier59.7$6.00
Qwen3.8 Maxfrontier58.1$3.00
Claude Opus 4.8frontier57.3$10.0
Muse Spark 1.2heavy56.8$2.00
GPT-5.6 Terraheavy56.6$4.50
Gemini 3.7 Flashheavy56.0$1.50
Grok 4.5frontier55.8$3.00
Claude Sonnet 5heavy55.3$4.00
DeepSeek V4 Proheavy53.2$1.98
Muse Spark 1.1heavy53.2$2.00
GLM-5.2heavy52.6$2.15
GPT-5.6 Lunamedium52.3$0.45
Gemini 3.5 Flashheavy52.0$3.38
DeepSeek V4 Flashmedium51.8$0.66
Gemini 3.6 Flashheavy51.6$3.00
Gemini 3.1 Profrontier47.7$4.50
Qwen3.7 Maxfrontier46.7$3.75
MiniMax M3frontier45.4$0.53
Kimi K2.6heavy45.1$1.71
Kimi K2.7 Codeheavy43.0$1.71
Inkling (xhigh)heavy42.3$1.76
Hunyuan Hy3heavy42.2$0.24
Inkling-Small (Preview)heavy41.2$0.53
Agnes 2.5 Pro Alphamedium39.7$0.56
Qwen3.7 Plusheavy39.4$0.70
MiniMax M2.7heavy38.9$0.53
Nemotron 3 Ultra 550B A55Bheavy38.3$1.14
Grok 4.3medium37.9$1.56
Ling 3.0 Flashmedium37.8$0.11
Gemini 3.5 Flash-Litemedium37.4$0.85
Claude Sonnet 4.6heavy36.8$6.00
Muse Glimmer 30Bmedium35.1$0.58
LongCat 2.0heavy34.0$1.30
Claude Haiku 4.5medium29.9$2.00
Gemini 3 Flashheavy27.9$1.12
Grok 4 Fastmedium27.9$0.28
Gemini 3.1 Flash Litemedium25.6$0.56
GPT-OSS 120Bmedium24.1$0.26
Nemotron 3.5 Lightning 30B A3Blight23.6$0.088
GLM-4.7 Flashlight23.3$0.15
Qwen3 VL 235Bmedium20.9$2.62
Llama 4 Maverickmedium14.5$0.41
Llama 4 Scoutlight10.3$0.30
Qwen3 30B A3Blight9.2$0.75

Price per weight class, over time

Median $/M blended per class — the tiers move independently, and the cheap tiers keep getting cheaper.

frontierheavymediumlight
12/0725/0731/0706/0812/08$0.30$1.00$3.00$10.0frontier $4.50heavy $1.76medium $0.56light $0.30

Median of list-price blends per class, from our own daily snapshots — recording started 12/07/2026, 30 snapshots so far. Log price axis; medians move when models enter or leave a class, not only when prices change.

Table view
Datefrontierheavymediumlight
12/07$10.0$3.38$0.37$0.15
13/07$10.0$3.38$0.37$0.15
14/07$10.0$3.38$0.37$0.15
22/07$6.00$2.15$0.37$0.15
23/07$6.00$2.15$0.37$0.15
24/07$6.00$2.00$0.37$0.15
25/07$6.00$2.00$0.37$0.15
26/07$6.00$2.00$0.56$0.30
27/07$6.00$2.00$0.85$0.30
28/07$6.00$2.00$0.85$0.30
29/07$6.00$2.00$0.85$0.30
30/07$6.00$2.00$0.85$0.30
31/07$6.00$1.71$0.56$0.30
01/08$6.00$1.71$0.56$0.30
02/08$6.00$1.71$0.56$0.30
03/08$6.00$1.71$0.56$0.30
04/08$6.00$1.71$0.56$0.30
05/08$6.00$1.71$0.56$0.30
06/08$6.00$1.76$0.45$0.30
07/08$6.00$1.71$0.56$0.30
08/08$6.00$1.71$0.56$0.30
09/08$6.00$1.71$0.56$0.30
10/08$6.00$1.71$0.56$0.30
11/08$6.00$1.71$0.56$0.30
12/08$6.00$1.71$0.56$0.30
13/08$4.50$1.71$0.56$0.30
14/08$4.50$1.76$0.56$0.30
15/08$4.50$1.76$0.56$0.30
16/08$4.50$1.76$0.56$0.30
17/08$4.50$1.76$0.56$0.30

The price of intelligence

What one point of the AA reasoning index costs — bought from the leader, or one step behind it.

the intelligence leaderbest value within 5 points of it
26/0730/0703/0807/0811/0815/08$0.05$0.10$0.15$0.20Claude Opus 5 $0.16/ptGrok 4.6 $0.049/pt

$/pt = $/M blended ÷ AA reasoning index. “Best value” = the cheapest model within 5 index points of that day's leader. AA capability history in our snapshots starts 26/07/2026. Definitions are ours; a cheaper weak model always has a lower ratio, which is why the comparison stays near the frontier.

Table view
DateLeader$/ptBest value$/pt
26/07Claude Opus 5$0.16Kimi K3$0.11
27/07Claude Opus 5$0.16Kimi K3$0.11
28/07Claude Opus 5$0.16Kimi K3$0.11
29/07Claude Opus 5$0.16Kimi K3$0.11
30/07Claude Opus 5$0.16Kimi K3$0.11
31/07Claude Opus 5$0.16Kimi K3$0.11
01/08Claude Opus 5$0.16Kimi K3$0.11
02/08Claude Opus 5$0.16Kimi K3$0.11
03/08Claude Opus 5$0.16Kimi K3$0.11
04/08Claude Opus 5$0.16Kimi K3$0.11
05/08Claude Opus 5$0.16Kimi K3$0.11
06/08Claude Opus 5$0.16Qwen3.8 Max$0.052
07/08Claude Opus 5$0.16Qwen3.8 Max$0.052
08/08Claude Opus 5$0.16Qwen3.8 Max$0.052
09/08Claude Opus 5$0.16Qwen3.8 Max$0.052
10/08Claude Opus 5$0.16Qwen3.8 Max$0.052
11/08Claude Opus 5$0.16Qwen3.8 Max$0.052
12/08Claude Opus 5$0.16Qwen3.8 Max$0.052
13/08Claude Opus 5$0.16Grok 4.6$0.049
14/08Claude Opus 5$0.16Grok 4.6$0.049
15/08Claude Opus 5$0.16Grok 4.6$0.049
16/08Claude Opus 5$0.16Grok 4.6$0.049
17/08Claude Opus 5$0.16Grok 4.6$0.049

Open weights vs proprietary

The best open-weights model sits 3.4 reasoning points behind the best proprietary one today.

best proprietarybest open weights
26/0730/0703/0807/0811/0815/08556065Claude Opus 5 63.1Kimi K3 59.7

Best AA reasoning index on each side, per snapshot (since 26/07/2026). The open/proprietary classification is ours — a vendor counts as open when it ships downloadable weights for the tracked model. The shaded band is the gap.

Table view
DateProprietaryAAOpenAAGap
26/07Claude Opus 560.7Kimi K357.13.6
27/07Claude Opus 560.7Kimi K357.13.6
28/07Claude Opus 560.7Kimi K357.13.6
29/07Claude Opus 560.7Kimi K357.13.6
30/07Claude Opus 560.7Kimi K357.13.6
31/07Claude Opus 560.7Kimi K357.13.6
01/08Claude Opus 560.7Kimi K357.13.6
02/08Claude Opus 560.7Kimi K357.13.6
03/08Claude Opus 560.7Kimi K357.13.6
04/08Claude Opus 560.7Kimi K357.13.6
05/08Claude Opus 560.7Kimi K357.13.6
06/08Claude Opus 563.1Kimi K359.73.4
07/08Claude Opus 563.1Kimi K359.73.4
08/08Claude Opus 563.1Kimi K359.73.4
09/08Claude Opus 563.1Kimi K359.73.4
10/08Claude Opus 563.1Kimi K359.73.4
11/08Claude Opus 563.1Kimi K359.73.4
12/08Claude Opus 563.1Kimi K359.73.4
13/08Claude Opus 563.1Kimi K359.73.4
14/08Claude Opus 563.1Kimi K359.73.4
15/08Claude Opus 563.1Kimi K359.73.4
16/08Claude Opus 563.1Kimi K359.73.4
17/08Claude Opus 563.1Kimi K359.73.4

Context window vs price

A long memory no longer commands a premium — million-token context now exists at every price tier. Bubble size = reasoning score.

256k512k1M2M$0.10$0.30$1.00$3.00$10.0context window →$/M blended (log)Claude Opus 5Grok 4 FastNemotron 3.5 Lightning 30B A3B

Both axes log. Context as the vendor lists it; price is the 3:1 blend. 4 tracked models publish no context figure and are not plotted.

Table view
ModelContext$/M blendedReasoning
Grok 4 Fast2,000k$0.2827.9
Claude Fable 51,000k$20.062.1
Claude Opus 4.81,000k$10.057.3
Claude Opus 51,000k$10.063.1
Claude Sonnet 4.61,000k$6.0036.8
Gemini 3.1 Pro1,000k$4.5047.7
Gemini 3 Flash1,000k$1.1227.9
Gemini 3.5 Flash1,000k$3.3852.0
GLM-5.21,000k$2.1552.6
DeepSeek V4 Flash1,000k$0.6651.8
Llama 4 Maverick1,000k$0.4114.5
Gemini 3.1 Flash Lite1,000k$0.5625.6
Claude Sonnet 51,000k$4.0055.3
MiniMax M31,000k$0.5345.4
DeepSeek V4 Pro1,000k$1.9853.2
Nemotron 3 Ultra 550B A55B1,000k$1.1438.3
Gemini 3.7 Flash1,000k$1.5056.0
Gemini 3.6 Flash1,000k$3.0051.6
Kimi K31,000k$6.0059.7
Muse Spark 1.11,000k$2.0053.2
Gemini 3.5 Flash-Lite1,000k$0.8537.4
Qwen3.7 Max1,000k$3.7546.7
Qwen3.8 Max1,000k$3.0058.1
Qwen3.7 Plus1,000k$0.7039.4
LongCat 2.01,000k$1.3034.0
Agnes 2.5 Pro Alpha1,000k$0.5639.7
Nemotron 3.5 Lightning 30B A3B1,000k$0.08823.6
Inkling (xhigh)524k$1.7642.3
Grok 4.6500k$3.0060.9
Grok 4.5500k$3.0055.8
GPT-5.6 Sol400k$11.260.9
GPT-5.6 Terra400k$4.5056.6
GPT-5.6 Luna400k$0.4552.3
Llama 4 Scout328k$0.3010.3
Qwen3 VL 235B262k$2.6220.9
Qwen3 30B A3B262k$0.759.2
Kimi K2.7 Code262k$1.7143.0
Kimi K2.6256k$1.7145.1
MiniMax M2.7205k$0.5338.9
GLM-4.7 Flash203k$0.1523.3
Claude Haiku 4.5200k$2.0029.9
GPT-OSS 120B131k$0.2624.1
Ling 3.0 Flash131k$0.1137.8
Muse Glimmer 30B131k$0.5835.1

The cache-discount landscape

What a cached input token costs as a share of a fresh one — the spread runs from 2% to 100%, and it decides who wins on repeated-context workloads like agents.

Agnes 2.5 Pro Alpha1%DeepSeek V4 Flash3%DeepSeek V4 Pro3%Claude Sonnet 4.610%GPT-5.6 Luna10%Gemini 3.5 Flash10%Gemini 3.7 Flash10%Gemini 3.6 Flash10%Kimi K310%Claude Fable 510%Claude Opus 4.810%Claude Opus 510%Claude Haiku 4.510%GPT-5.6 Terra10%Gemini 3.1 Pro10%Gemini 3 Flash10%Gemini 3.1 Flash Lite10%Claude Sonnet 510%Qwen3.7 Max10%Muse Glimmer 30B11%Muse Spark 1.112%Qwen3.8 Max12%

Cache READ price ÷ input price, list prices. 15 more models in the table; 12 tracked models publish no cache price. Cache WRITE surcharges (where they exist) are not in this ratio.

Table view
ModelCached shareCached $/MInput $/M
Agnes 2.5 Pro Alpha1%$0.004$0.45
DeepSeek V4 Flash3%$0.014$0.44
DeepSeek V4 Pro3%$0.044$1.32
Claude Sonnet 4.610%$0.30$3.00
GPT-5.6 Luna10%$0.02$0.20
Gemini 3.5 Flash10%$0.15$1.50
Gemini 3.7 Flash10%$0.075$0.75
Gemini 3.6 Flash10%$0.15$1.50
Kimi K310%$0.30$3.00
Claude Fable 510%$1.00$10.0
Claude Opus 4.810%$0.50$5.00
Claude Opus 510%$0.50$5.00
Claude Haiku 4.510%$0.10$1.00
GPT-5.6 Terra10%$0.20$2.00
Gemini 3.1 Pro10%$0.20$2.00
Gemini 3 Flash10%$0.05$0.50
Gemini 3.1 Flash Lite10%$0.025$0.25
Claude Sonnet 510%$0.20$2.00
Qwen3.7 Max10%$0.25$2.50
Muse Glimmer 30B11%$0.04$0.35
Muse Spark 1.112%$0.15$1.25
Qwen3.8 Max12%$0.25$2.00
GLM-4.7 Flash14%$0.01$0.07
Qwen3 VL 235B14%$0.10$0.70
Kimi K2.615%$0.14$0.95
Grok 4.316%$0.20$1.25
Inkling (xhigh)17%$0.17$1.00
GLM-5.219%$0.26$1.35
MiniMax M2.720%$0.06$0.30
Kimi K2.7 Code20%$0.19$0.95
MiniMax M320%$0.06$0.30
Grok 4.625%$0.50$2.00
Grok 4.525%$0.50$2.00
GPT-OSS 120B27%$0.04$0.15
Hunyuan Hy327%$0.037$0.14
Qwen3 30B A3B50%$0.10$0.20
Llama 4 Maverick63%$0.17$0.27

Release cadence

24 price-relevant models entered the tracker since 13/07 — the cadence problem is real: this strip is what “keeping up” actually looks like.

Google (3)Meta (3)Alibaba (3)Anthropic (2)Moonshot AI (2)Thinking Machines (2)NVIDIA (2)MiniMax (1)DeepSeek (1)Tencent (1)Meituan (1)InclusionAI (1)Agnes AI (1)xAI (1)13/0717/0801/08

A dot = the day a model entered OUR tracking, which for post-launch additions is within days of its release — not the vendor's launch date. The 26 models already tracked on 12/07 (day one) are not shown.

Table view
DateVendorModel
13/08xAIGrok 4.6
13/08GoogleGemini 3.7 Flash
12/08NVIDIANemotron 3.5 Lightning 30B A3B
11/08MetaMuse Glimmer 30B
07/08NVIDIANemotron 3 Ultra 550B A55B
07/08Agnes AIAgnes 2.5 Pro Alpha
06/08MetaMuse Spark 1.2
06/08InclusionAILing 3.0 Flash
06/08AlibabaQwen3.8 Max
31/07Thinking MachinesInkling-Small (Preview)
30/07MeituanLongCat 2.0
25/07AnthropicClaude Opus 5
24/07TencentHunyuan Hy3
24/07AlibabaQwen3.7 Plus
24/07AlibabaQwen3.7 Max
23/07Thinking MachinesInkling (xhigh)
23/07GoogleGemini 3.5 Flash-Lite
22/07Moonshot AIKimi K3
22/07MetaMuse Spark 1.1
22/07GoogleGemini 3.6 Flash
14/07MiniMaxMiniMax M3
14/07DeepSeekDeepSeek V4 Pro
13/07Moonshot AIKimi K2.7 Code
13/07AnthropicClaude Sonnet 5