August 9 scrape of Artificial Analysis shows Claude Opus 5 still #1, but the whole frontier band rebased upward: Opus 5 max 63 (July issue 61), Fable 5 with fallback 62, GPT-5.6 Sol max 61, open-weights Kimi K3 max 60. Muse Spark 1.2 (xhigh) enters at 57; Grok 4.5 high sits at 56.

Current leader

Claude Opus 5

Anthropic

Still #1 on AA Intelligence Index — now 63 max, up from 61 in the July issue.

Score

01Intelligence Index: 63 (max)
02SWE-bench Verified ~96%
03$5 / $25 per 1M tokens
041M context
05Released 2026-07-24

Runners-up

№2

Claude Fable 5

Anthropic

AA lists Fable 5 with fallback at 62 — #2 overall after the August rebase.

Intelligence Index: 62 (with fallback)
Mythos-class safeguards
1M+ context · Adaptive Reasoning
Released 2026-06-09

№3

GPT-5.6 Sol

OpenAI

Sol max climbs to 61 on AA; still the OpenAI flagship tier.

Intelligence Index: 61 (max)
Terminal-Bench 2.1: 88.8%
$5 / $30 per 1M tokens
1M context
GA 2026-07-09

№4

Kimi K3

Moonshot AI

Open-weights K3 max hits 60 — global #4, only 3 points off Opus 5.

Intelligence Index: 60 (open #1)
2.8T MoE · 1M context
LMArena WebDev #1 (July)
Open weights · $3/$15 per 1M
Released 2026-07-17

№5

Muse Spark 1.2

Muse

New mid-frontier entrant at AA Index 57 (xhigh) — displaces Grok from the July top-5 shape.

Intelligence Index: 57 (xhigh)
New vs July top tier
Closed-weights contender

№6

Grok 4.5

xAI

AA high mode at 56; still the cheap always-on reasoning pick at $2/$6.

Intelligence Index: 56 (high)
Aggressive pricing: $2 / $6 per 1M
Always-on reasoning · 1M context
Released 2026-07-08

Change

AA Intelligence Index top-4 all +2–3 vs July issue. Muse Spark 1.2 is new in the mid-frontier; DeepSeek V4 Flash 0731 lands at 52.

Market

Opus 5 remains the default frontier workhorse at $5/$25; Fable 5 is the Mythos-class ceiling; Sol is OpenAI’s tiered answer; Kimi K3 is still the open model inside the frontier band.

Image Generation

GPT Image 2 (high) extends its Artificial Analysis Image Arena lead to ELO 1357 (July issue 1338). Reve 2.1 holds #2 at 1314. Google’s Nano Banana 2 (Gemini 3.1 Flash Image Preview) debuts in the top 3 at 1307. Seedream 5.0 Pro sits at 1266.

Current leader

GPT Image 2

OpenAI

AA Image Arena ELO 1357 (high) — quality and ecosystem crown extended.

Score

1357

01AA T2I Arena ELO: 1357 (#1)
02Token-based pricing
03Batch API at 50% discount
04High-fidelity inputs

Runners-up

№2

Reve 2.1

Reve

Holds #2 at ELO 1314 with layered generation.

AA T2I Arena ELO: 1314 (#2)
Layered generation
$24 per 1k images
Day-one top-tier quality

1314

№3

Nano Banana 2

Google

Gemini 3.1 Flash Image Preview branding — new AA top-3 entrant at 1307.

AA T2I Arena ELO: 1307 (#3)
Gemini 3.1 Flash Image Preview
Google stack integration

1307

№4

MAI-Image-2.5

Microsoft

Enterprise image default at ELO 1298.

AA T2I Arena ELO: 1298
Enterprise integration
Microsoft stack

1298

№5

Seedream 5.0 Pro

ByteDance

Top Chinese production pick at ELO 1266 with strong text rendering.

AA T2I Arena ELO: 1266
Layer separation
Strong text rendering

1266

Change

GPT Image 2 high 1338→1357; Nano Banana 2 enters top 3; Seedream 5.0 Pro still the leading Chinese production option in the top tier.

Market

Quality + ecosystem: GPT Image 2; layered generation value: Reve 2.1; Google preview stack: Nano Banana 2; Chinese text/layout: Seedream 5.0 Pro.

Video Generation

Gemini Omni Flash keeps the AA text-to-video (with audio) crown at ELO 1244 (July 1240). MiniMax H3 surges to #2 at 1240. Dreamina Seedance 2.0 720p is #3 at 1224; Wan2.7-260612 and HappyHorse-1.1 round out the top 5. Sora API sunset remains 2026-09-24.

Current leader

Gemini Omni Flash

Google

AA T2V with audio #1 at ELO 1244 — still the synced A/V crown.

Score

1244

01AA T2V Arena (w/ audio) #1 · ELO 1244
02Without-audio arena also #1 · ELO 1324
03Native audio-visual sync
04Gemini Omni stack

Runners-up

№2

MiniMax H3

MiniMax

New #2 on AA T2V with audio at ELO 1240 — the biggest August video mover.

AA T2V Arena (w/ audio) #2 · ELO 1240
Without-audio #2 · ELO 1306
Fast climb vs July field

1240

№3

Seedance 2.0

ByteDance

Dreamina Seedance 2.0 720p at ELO 1224 — still the ByteDance production workhorse.

AA T2V Arena (w/ audio) #3 · ELO 1224
Production pipeline maturity
Multimodal references

1224

№4

Wan2.7-260612

Alibaba

Alibaba Wan2.7 build 260612 at ELO 1161 on the with-audio arena.

AA T2V Arena (w/ audio) ELO: 1161
Alibaba Cloud native

1161

№5

Kling 3.0 Pro

Kuaishou 快手

Still the APAC short-form value line (July ELO band); not displaced by H3/Seedance fight at the very top.

Native 4K/60fps line
Turbo iteration speed
TikTok / Douyin native style

1110

Change

With-audio arena: Gemini Omni Flash 1240→1244; MiniMax H3 takes #2 from Seedance; Seedance 2.0 still #3.

Market

Synced A/V arena: Gemini Omni Flash; aggressive #2 challenger: MiniMax H3; ByteDance production: Seedance 2.0; APAC short-form still Kling-family adjacent.

Code Generation & Agentic Coding

Claude Opus 5 remains the agentic-coding leader on the July SWE-bench Verified high-water mark (~96%). Kimi K3 keeps the open-weights / LMArena WebDev crowd-vote crown (1679). GPT-5.6 Sol still leads Terminal-Bench 2.1 at 88.8%. No cleaner public number beating Opus 5 was confirmed on the August 9 pass.

Current leader

Claude Opus 5

Anthropic

SWE-bench Verified ~96% high-water mark still stands; AA Index now 63.

Score

01SWE-bench Verified: ~96% (#1)
02AA Intelligence Index: 63
03$5 / $25 per 1M tokens
04Long-horizon agentic edits

Runners-up

№2

Kimi K3

Moonshot AI

Open-weights WebDev arena king at 1679; AA Index 60.

LMArena WebDev: 1679 (#1)
AA Intelligence Index: 60
Open weights · 2.8T MoE
$3 / $15 per 1M tokens

1679

№3

Claude Fable 5

Anthropic

Mythos-class coding ceiling; AA Index 62 overall.

SWE-bench Verified: 95%
Mythos-class agentic coding
1M+ context

№4

GPT-5.6 Sol

OpenAI

Best sandboxed terminal agent — Terminal-Bench 2.1 88.8%.

Terminal-Bench 2.1: 88.8% (#1)
SWE-bench Pro: 64.6%
Sandboxed shell built-in

88.8

Change

No crown flip: Opus 5 still SWE-bench leader; Kimi K3 still WebDev #1; AA overall Index now 63/60 for Opus/Kimi.

Market

Hardest refactors: Opus 5; open frontend + deploy: Kimi K3; Mythos ceiling: Fable 5; sandboxed terminal agents: GPT-5.6 Sol.

Voice / Speech

No new speech-to-speech king — OpenAI Realtime 2 still leads agentic voice. On AA TTS, Qwen-Audio-3.0-TTS-Plus remains #1 at ELO 1229 (July 1234; same crown, slight Elo drift). Speechify Simba 3.2 is #2 at 1227; Gemini 3.1 Flash TTS #3 at 1210.

Current leader

Realtime 2

OpenAI

Still the agentic voice default — configurable-reasoning speech-to-speech.

01Configurable reasoning
02Speech-to-speech agents
03Streaming translate + STT variants
04Released 2026-05-07

Runners-up

№2

Qwen-Audio-3.0-TTS-Plus

Alibaba

AA TTS #1 at ELO 1229 — China still holds the component TTS crown.

AA TTS ELO: 1229 (#1)
Strong multilingual + CJK
Alibaba Cloud native

1229

№3

Simba 3.2

Speechify

AA TTS #2 at ELO 1227 — inches behind Qwen.

AA TTS ELO: 1227 (#2)
Consumer reading voice quality

1227

№4

ElevenLabs v3

ElevenLabs

Still the character voice cloning default; Scribe v2 realtime STT remains in stack.

Character voice cloning SOTA
Scribe v2 realtime STT
Music v2 API
100+ languages

Change

TTS crown unchanged (Qwen-Audio-3.0-TTS-Plus); Elo 1234→1229. Agentic S2S still Realtime 2.

Market

Agentic voice: Realtime 2; raw TTS quality: Qwen-Audio-3.0-TTS-Plus; character cloning: ElevenLabs v3.

Music Generation

Suno v6 still has not shipped as of the August 9 check. Suno v5.5 remains the full-song default. ElevenLabs Music v2 (API since June 15) and Udio v3 stems stay the secondary studio picks; Lyria remains on the Gemini Omni cross-modal path.

Current leader

Suno v5.5

Suno

Still the latest shipped Suno — best full-song coherence and lyric prosody.

01Full-song coherence SOTA
02Lyric prosody best
03Multilingual vocal
04Style transfer

Runners-up

№2

Udio v3

Udio

Studio-grade mixing with stem-level export.

Stem-level export
Studio-grade mixing
Strong electronic genres
DAW-friendly workflow

№3

ElevenLabs Music v2

ElevenLabs

API-first licensed-catalog music generation since June 15.

API-first music generation
Licensed catalog
Music Finetunes API
Voice stack integration

Change

No crown change. Suno v6 still unannounced/unshipped.

Market

Full songs: Suno v5.5; stems/studio: Udio v3; licensed API music: ElevenLabs Music v2; cross-modal: Lyria via Gemini Omni.

Vision / Multimodal Understanding

Claude Fable 5 remains the LMArena Vision leader at ELO 1327 from the July recalculation — no higher public Vision number was confirmed on August 9. Overall AA Intelligence still has Opus 5 at 63 and Fable 5 at 62, so multimodal understanding stays Anthropic-heavy. Gemini 3.6 Flash (Index 52) is the fast Google multimodal pick.

Current leader

Claude Fable 5

Anthropic

LMArena Vision #1 at 1327; AA overall Index 62.

Score

1327

01LMArena Vision ELO: 1327 (#1)
02AA Intelligence Index: 62
03OCR + document SOTA
041M+ context

Runners-up

№2

Claude Opus 5

Anthropic

Multimodal flagship with AA overall #1 at 63.

AA Intelligence Index: 63 (#1 overall)
Document Q&A + charts
$5 / $25 per 1M tokens

№3

GPT-5.6 Sol

OpenAI

Image-grounded reasoning at AA Index 61.

AA Intelligence Index: 61
Image-grounded reasoning
1M context

№4

Gemini 3.6 Flash

Google

New on AA overall at 52 — fast Google multimodal stack pick.

AA Intelligence Index: 52
High output speed on AA
Gemini multimodal stack

Change

Vision crown unchanged (Fable 5 @ 1327). Gemini 3.6 Flash appears on AA overall at 52.

Market

OCR/docs/charts: Anthropic; image-grounded reasoning: GPT-5.6 Sol; video understanding: Gemini line.

Open-Source / Open-Weights

Kimi K3 max reaches AA Intelligence Index 60 — still open #1 and global #4, only 3 points behind Opus 5 (63). DeepSeek V4 Flash 0731 jumps to 52 without a price change ($0.14/$0.28). GLM-5.2 max is ~53; MiniMax-M3 is 45.

Current leader

Kimi K3

Moonshot AI

Open #1 at AA Index 60 — global #4, 3 points off Opus 5.

Score

01AA Intelligence Index: 60 (open #1)
022.8T MoE · 1M context
03LMArena WebDev #1
04$3 / $15 per 1M tokens

Runners-up

№2

DeepSeek V4 Flash 0731

DeepSeek

0731 refresh to Index 52 at unchanged $0.14/$0.28 — open value king.

AA Intelligence Index: 52
$0.14 in / $0.28 out per 1M
MIT open weights · 1M context

№3

GLM-5.2

Zhipu AI

AA max around 53 — strong open Chinese alternative.

AA Intelligence Index: ~53 (max)
Pay-as-you-go API available

№4

Gemma 4 12B

Google

Apache 2.0 on-device multimodal default.

Apache 2.0 · open weights
256K context · native multimodal
Runs on 16GB VRAM

Change

Kimi K3 57→60; DeepSeek V4 Flash 0731 47→52; open crown unchanged.

Market

Frontier open deploy: Kimi K3; pure $/IQ: DeepSeek V4 Flash 0731; on-device multimodal: Gemma 4 12B.

Intelligence per Dollar

Cost-Effectiveness / Value

DeepSeek V4 Flash 0731 is the new value headline: Intelligence Index 52 at the same $0.14/$0.28 pricing that made Flash the July king at 47. MiniMax-M3 remains the cheap high-coding tier at Index 45. Kimi K3 offers Index 60 but at $3/$15 — quality, not pure $/IQ.

Previously: DeepSeek V4 Flash

Current leader

DeepSeek V4 Flash 0731

DeepSeek

Index 52 at $0.14/$0.28 — still the intelligence-per-dollar king after the 0731 refresh.

Score

01AA Intelligence Index: 52
02$0.14 in / $0.28 out per 1M
03Blended cost still ~1/10 of flagships
04MIT open weights · 1M context

Runners-up

№2

MiniMax-M3

MiniMax

AA Index 45 — cheap high-coding value tier.

AA Intelligence Index: 45
Strong coding-per-dollar
Standard API pricing tier

№3

GLM-5.2

Zhipu AI

Higher absolute price than Flash, stronger index (~53 max).

AA Intelligence Index: ~53 (max)
Pay-as-you-go API

№4

Kimi K3

Moonshot AI

Not the cheapest — but best open quality at $3/$15 with Index 60.

AA Intelligence Index: 60
$3 / $15 per 1M tokens
Open weights frontier-adjacent

Change

Leader name updates to V4 Flash 0731; Index 47→52 with unchanged price. MiniMax-M3 still the SWE-bench value pick from July narrative.

Market

Highest volume low-cost: V4 Flash 0731; stronger open reasoning on a budget: Kimi K2.6-class / GLM; quality-per-dollar frontier-adjacent: Kimi K3 if you can pay $15 out.

Editorial · 05 observations

What changed this month

What changed across the AI model landscape this month — distilled from the data above.

Frontier scores rebased upward

Artificial Analysis Intelligence Index top-4 all rose ~2–3 points vs the July issue (Opus 63, Fable 62, Sol 61, Kimi 60), without a crown flip at the very top.

DeepSeek Flash 0731 value jump

DeepSeek V4 Flash 0731 reaches Index 52 at unchanged $0.14/$0.28 pricing — the open cheap tier got smarter without getting pricier.

MiniMax H3 crashes the video podium

MiniMax H3 takes #2 on AA text-to-video with audio (ELO 1240), slotting between Gemini Omni Flash (1244) and Seedance 2.0 (1224).

Google Nano Banana 2 enters image top-3

Nano Banana 2 (Gemini 3.1 Flash Image Preview) debuts at AA Image Arena ELO 1307, behind only GPT Image 2 and Reve 2.1.

Sora sunset still on the clock

OpenAI’s Sora API shutdown date remains 2026-09-24; video demand continues consolidating around Gemini Omni, MiniMax H3, and Seedance.

Sources

[01]
Artificial Analysisbenchmark
2026-08-09
[02]
Artificial Analysis Text-to-Imagebenchmark
2026-08-09
[03]
Artificial Analysis Video Arenabenchmark
2026-08-09
[04]
Artificial Analysis Speech Leaderboardsbenchmark
2026-08-09
[05]
SWE-bench Leaderboardbenchmark
2026-08-09
[06]
OpenAI Changelogofficial changelog
2026-08-09
[07]
Anthropic Newsofficial changelog
2026-08-09
[08]
DeepSeek API Pricingofficial changelog
2026-08-09
[09]
Moonshot AI / Kimiofficial changelog
2026-08-09