Hedge

AI & Technology Update

Frontier models · Model comparison · Markets · Capital

Saturday, 12 September 2026 · Hedge Research Desk

The Read

AI safety warnings from industry insiders are prompting regulatory scrutiny and market debate, even as leading labs rapidly deploy advanced models like GPT-6 Astra and expand enterprise tools.

Strategy · Execution · Growth

Claude Fable 597/100Claude Opus 4.893/100Kimi K2.7-Code93/100OpenAIOpenAIOpenAIFunding$500MFunding$100MFunding$1.5BClaude Fable 597/100Claude Opus 4.893/100Kimi K2.7-Code93/100OpenAIOpenAIOpenAIFunding$500MFunding$100MFunding$1.5B

Nvidia (NVDA)

$218.29

-4.45% today

Microsoft (MSFT)

$495.63

-2.84% today

Alphabet (GOOGL)

$338.50

-1.16% today

Meta (META)

$648.03

+6.12% today

Top coder

97/100

Claude Fable 5

Funding

$500M

deal

Frontier Models

The week across the frontier labs — new defaults, previews and open-weight releases — plus where the coding benchmarks stand.

OpenAI

GPT-6 Astra — (14 Sep) Perplexity is utilizing GPT-6 Astra to write communications, modify software, and monitor production systems with less frequent human oversight than previous models. · 14 Sep

Read → OpenAI

OpenAI

ChatGPT for Financial Services — (11 Sep) OpenAI launched a specialized ChatGPT product for the financial sector, targeting the research, modeling, and pitchbook tasks typically handled by junior bankers. · 11 Sep

Read → CNBC

OpenAI

GPT-6 Astra — (11 Sep) Cognition is integrating GPT-6 Astra into its Devin software engineering agent to improve automated testing and code verification workflows. · 11 Sep

Read → OpenAI

OpenAI

ChatGPT Pro — (10 Sep) OpenAI has temporarily paused sign-ups for its Pro subscription tier due to high demand straining its systems following the rollout of GPT-6 Astra. · 10 Sep

Read → TechCrunch

Meta

Muse — (10 Sep) Meta's newly launched AI agent application, Muse, has reached the second most downloaded app position in the United States. · 10 Sep

Read → TechCrunch

Coding LeaderboardCoding score — Hedge Model Comparison V3.3 (maintained dataset)

Claude Fable 5 Anthropic

97/100

Claude Opus 4.8 Anthropic

93/100

Kimi K2.7-Code Moonshot

93/100

GPT-5.6 Sol OpenAI

92/100

Kimi K2.6 Moonshot

90/100

GLM-5V-Turbo (vision) Zhipu

90/100

Claude Sonnet 5 Anthropic

89/100

GLM-5.2 Zhipu

88/100

Model Comparison

The full AI Models Comparison dashboard — every category, factor weights, best-for and sources. Benchmark-anchored; ✅ verified · ⚠️ partial · ❓ released, not yet independently benchmarked. Wide tables scroll sideways.

V3.3 · benchmark-anchored · trimmed to the latest model generation (6 Jul 2026). Sub-scores are analytical mappings from public benchmarks — not vendor claims.

LLMs

Top score

Claude Fable 5 — SWE-V 95%

New OpenAI flagship

GPT-5.6 Sol — AA Index 59, coding #1

Open-weights leader

GLM-5.2

#1 open on AA Index (51) · SWE-Pro 62.1

Cheapest viable

$0.14

DeepSeek V4-Flash /1M in · SWE-V 79%

V3.3 — trimmed to the current generation

Removed old-generation tiers (Claude Haiku 4.5, Gemini 3.1 Pro / 3.1 Flash-Lite) — each is still that maker’s current model in the tier, so dropped for a latest-generation board, not deprecated. Gemini’s LLM entry is now 3.5 Flash (Pro tier temporarily empty, delayed ~17 Jul). Current Anthropic line: Fable 5 · Opus 4.8 · Sonnet 5.

Show score ≥

#	Model	Score	Business use	Web research	Presentation	Low halluc.	Coding	Tool use	Docs	Cost (in/out $)	Open?	Conf.
1	Claude Fable 5	93	● 95	● 90	● 93	● 96	● 97	● 95	● 94	$10 / $50	No	✅
2	Claude Opus 4.8	90	● 93	● 82	● 91	● 92	● 93	● 92	● 93	$5 / $25	No	✅
3	GPT-5.6 Sol	89	● 90	● 90	● 88	● 82	● 92	● 92	● 90	$5 / $30	No	✅
4	Claude Sonnet 5	86	● 90	● 80	● 88	● 89	● 89	● 90	● 90	$3 / $15*	No	✅
5	Kimi K2.6	86	● 88	● 92	● 88	● 95	● 90	● 94	● 88	$0.95 / $4.00	Yes	✅
6	GLM-5.2	83	● 78	● 78	● 76	● 80	● 88	● 86	● 76	$1.40 / $4.40	Yes	✅
7	Qwen 3.7-Max	83	● 80	● 74	● 78	● 78	● 86	● 88	● 76	$1.25 / $3.75*	No	✅
8	GPT-5.6 Terra	82	● 84	● 84	● 80	● 80	● 86	● 85	● 84	$2.50 / $15.00	No	✅
9	Grok 4.5	82	● 78	● 82	● 76	● 78	● 86	● 84	● 76	$2 / $6	No	✅
10	Kimi K2.7-Code	82	● 78	● 80	● 76	● 90	● 93	● 90	● 78	$0.95 / $4.00	Yes	⚠️
11	GLM-5V-Turbo (vision)	81	● 78	● 76	● 82	● 76	● 90	● 86	● 82	$1.20 / $4.00	No	⚠️
12	Gemini 3.5 Flash	81	● 76	● 88	● 78	● 80	● 85	● 82	● 76	$1.50 / $9.00	No	✅
13	DeepSeek V4-Pro	81	● 78	● 75	● 76	● 82	● 86	● 80	● 78	$0.44 / $0.87	Yes	⚠️
14	MiniMax M3	78	● 75	● 88	● 76	● 75	● 82	● 80	● 78	$0.30 / $1.20	Yes	⚠️
15	DeepSeek V4-Flash	75	● 70	● 68	● 68	● 74	● 82	● 74	● 70	$0.14 / $0.28	Yes	✅
16	MiMo-V2.5-Pro	73	● 72	● 70	● 72	● 74	● 80	● 76	● 74	$0.44 / $0.87	No	✅
17	GPT-5.6 Luna	72	● 72	● 74	● 70	● 76	● 74	● 74	● 72	$1 / $6	No	✅
18	ERNIE 5.1	72	● 74	● 72	● 72	● 74	● 74	● 72	● 74	$0.59 / $2.65	No	✅
19	Qwen 3.7-Plus	72	● 74	● 72	● 74	● 76	● 82	● 76	● 74	$0.32 / $1.28	No	⚠️

Cost = list API price, input / output per 1M tokens. ● colour: green ≥85 · amber 72–84 · grey 60–71 · red <60. Open? = open-weights availability.

Voice AI

Best Egyptian S2S

gpt-realtime-1.5

Grok Voice Agent close 2nd, 3× cheaper

Cheapest real-time S2S

$0.008/min

Gemini 3.1 Flash Live

Egyptian TTS confirmed

ElevenLabs v3

Only vendor confirming EG dialect

Cheapest Arabic STT

$0.10/hr

Grok STT — diarization + timestamps

Egyptian Arabic gate

Voice models are judged first on Egyptian Arabic quality (the top two weighted factors). Only ElevenLabs v3 explicitly confirms Egyptian dialect; others list generic "Arabic" and are flagged ⚠️. GLM-4-Voice is Chinese/English only and fails the Egyptian gate.

S2S — full real-time conversation (5 models)

Show score ≥

#	Model	Score	Egyptian quality	Naturalness	Emotion	Cloning	Real-time	$/min audio	Conf.
1	gpt-realtime-1.5	90	● 95	● 92	● 85	● 70	● 85	~$0.18–0.46/min	⚠️
2	Grok Voice Agent	88	● 95	● 90	● 85	● 88	● 95	$0.05/min · 2-min clone	✅
3	gpt-realtime-mini	80	● 92	● 85	● 85	● 60	● 90	$0.063–0.146/min	❓
4	Gemini 3.1 Flash Live	78	● 82	● 86	● 78	● 0	● 92	~$0.008/min	⚠️
5	GLM-4-Voice	55	● 30	● 80	● 75	● 60	● 90	Open · CN/EN only	⚠️

TTS — text-to-speech (6 models)

Show score ≥

#	Model	Score	Egyptian quality	Naturalness	Emotion	Cloning	Latency	$/min audio	Conf.
1	ElevenLabs v3 Multilingual	88	● 92	● 95	● 95	● 95	● 80	$0.10–0.30/1K chars · Egyptian ✅	✅
2	Grok TTS	86	● 90	● 88	● 80	● 85	● 92	$4.20/1M chars · 120s clone	✅
3	ElevenLabs Flash / Turbo	78	● 85	● 88	● 80	● 90	● 95	$0.05/1K chars · ~75ms	✅
4	Microsoft MAI-Voice-2	70	● 60	● 85	● 78	● 0	● 80	$22/1M chars · Arabic ⚠️	⚠️
5	GLM-TTS	68	● 55	● 85	● 80	● 88	● 85	Open · 3s clone	⚠️
6	Mistral Voxtral TTS	66	● 55	● 82	● 72	● 0	● 80	$0.016/1K chars	⚠️

STT — speech-to-text (6 models)

Show score ≥

#	Model	Score	Egyptian / Arabic	Languages	Real-time	Accuracy	$/min audio	Conf.
1	Grok STT	82	● 85	25	● 90	● 92	$0.10/hr batch · diarization	✅
2	GLM-ASR-2512	76	● 80	30	● 90	● 88	Open · Arabic ✅	⚠️
3	ElevenLabs Scribe v2	76	● 78	90	● 90	● 91	$0.39/hr · 150ms	❓
4	OpenAI gpt-4o-transcribe	74	● 70	99	● 85	● 90	$0.003/min mini	⚠️
5	gpt-realtime-whisper	72	● 70	99	● 92	● 88	$0.017/min streaming	⚠️
6	Gemini 3.1 Flash STT	70	● 68	90	● 88	● 86	~$0.005–0.01/min	⚠️

$/min normalised from list prices using ~500 chars/min & ~1,500 tok/min speech-rate assumptions. Open (self-host) = weights on Hugging Face, no API fee.

Image Gen

Top score

GPT Image 2 — Arena Elo 1339

Best Arabic rendering

GPT Image 2

Pixel-perfect Arabic (only confirmed premium)

Best open + Arabic

HunyuanImage 3.0

80B MoE, multilingual, self-hostable

Cheapest viable

$0.002

Luma Photon Flash /image (1080p)

Arabic text-in-image is the gate for Hedge / Arobah marketing

Score = composite of Arena Elo (25%), text accuracy (18%), Arabic rendering (15%), prompt adherence (12%), cost (10%), speed (8%), API reliability (7%), resolution (5%). DALL-E 3 retired 12 May 2026 — migrate to GPT Image 2. Arabic rendering independently confirmed only for GPT Image 2, Qwen-Image 2.0 and HunyuanImage 3.0; other Arabic sub-scores are conservative estimates (⚠️).

Show score ≥

#	Model	Maker	Score	Arena Elo	Text in image	Arabic text	Prompt adh.	Speed	$/image	Open?	Conf.
1	GPT Image 2 (high)	OpenAI	93	1339	● 99	● 95	● 92	● 75	$0.211 (high) · $0.053 (med)	No	✅
2	Reve 2.0	Reve	84	1281	● 88	● 60	● 88	● 82	$0.024 · $0.04 edit	No	⚠️
3	MAI-Image-2.5	Microsoft	83	1271	● 85	● 60	● 86	● 80	~$0.02–0.06 (token)	No	⚠️
4	GPT Image 1.5 (high)	OpenAI	82	1260	● 90	● 82	● 86	● 75	$0.16 (high)	No	✅
5	Qwen-Image 2.0	Alibaba	82	1180	● 92	● 88	● 88	● 78	~$0.05 (varies)	Yes	✅
6	HiDream-O1-Image 1.5	HiDream	81	1265	● 84	● 58	● 88	● 80	credit-based	Yes	⚠️
7	Cosmos3-Super	NVIDIA	80	1227	● 78	● 55	● 80	● 78	Free (open, self-host)	Yes	⚠️
8	Gemini 3.1 Flash Image	Google	80	1170	● 85	● 70	● 86	● 90	$0.067 (1K) · $0.101 (2K)	No	✅
9	HunyuanImage 3.0	Tencent	80	1152	● 88	● 85	● 88	● 80	Free (open 80B MoE)	Yes	✅
10	Ideogram 4.0	Ideogram	79	1200	● 96	● 60	● 90	● 82	$0.03 · $0.06 · $0.10	Partial	⚠️
11	Imagen 4 Ultra	Google	79	1150	● 80	● 68	● 88	● 60	$0.06 · $0.02 Fast	No	⚠️
12	Seedream 5.0 Lite	ByteDance	78	1150	● 80	● 58	● 84	● 88	$0.031 + $0.007 web	No	⚠️
13	CogView-4	Zhipu	78	1180	● 82	● 45	● 88	● 80	$0.03 · Apache 2.0	Yes	⚠️
14	FLUX.2 Pro	Black Forest Labs	77	1140	● 80	● 55	● 86	● 85	$0.03–0.073	Partial	✅
15	Grok Imagine 1.5	xAI	72	1173	● 70	● 55	● 78	● 85	$0.02	No	❓
16	Luma Photon	Luma Labs	70	1100	● 72	● 50	● 82	● 92	$0.015 · $0.002 Flash	No	❓

$/image = representative tier. Open? Yes = weights downloadable (licence varies) · Partial = restricted licence · No = API only.

Video Gen

Video generation — GLM-seeded (provisional)

Currently holds Zhipu/GLM video models only. Scores are provisional (relative VBench-2.0 standing) and flagged ⚠️. To make this a real cross-maker comparison, add peers (Sora 2, Veo 3, Kling, Runway Gen-4, Alibaba Wan) — say the word and they get researched + populated. No Arabic-prompt testing is published for any video model.

Show score ≥

#	Model	Maker	Score	Max resolution	Max duration	$/clip	Open?	Conf.
1	CogVideoX-1.5	Zhipu	72	768×1360	~10s @ 16fps	Open (self-host)	Yes	⚠️
2	CogVideoX	Zhipu	70	1440×960	~30s	Open (self-host)	Yes	✅
3	Ying (Qingying)	Zhipu	68	1440×960	~6s clips	CN market · n/a	No	⚠️

Open (self-host) = weights on Hugging Face; run on your own GPUs, no per-clip API fee.

Factor weights

LLM weights (sums to 100%)

Factor	Weight %	Tier
Hallucination level (low = better)	16	critical
Following instructions exactly	11	high
Document generation quality	9	high
Long task stamina	9	high
Tool use / agent reliability	8	high
Coding quality — real software	7	medium
Skills / tools per conversation	7	medium
Research & web browsing	7	medium
Hard reasoning / problem solving	5	medium
Image / chart understanding	5	medium
Cost — input + output	4	lower
Knowledge depth (PhD-level)	4	lower
Frontend / web design	3	lower
Coding — long projects	3	lower
Math accuracy	1	lower
Memory — claimed max	1	lower

Voice AI weights

Factor	Weight %	Tier
How natural it sounds (Egyptian)	22	critical
Pronunciation accuracy (Egyptian)	18	critical
Emotion & expressiveness	12	high
Voice cloning quality	12	high
Real-time conversation (S2S)	10	high
Steerability (tone control)	8	medium
Speed to first word	7	medium
Speech-to-text included	5	medium
Cost	4	lower
Voice variety	2	lower

Best for…

LLMs

Top reasoning + coding (any class)

Claude Fable 5

SWE-Bench Verified 95% and Pro 80.3% — both #1 by a clear margin. HLE-with-tools 64.5, AA Index 60. Premium ($10/$50).

New OpenAI flagship (agentic coding)

GPT-5.6 Sol

GA 9 Jul. AA Index 59, Coding Agent Index 80 (top of the agentic table), Terminal-Bench 88.8, GPQA 89. $5/$30.

Best value at scale (any class)

Kimi K2.6

Score 86 at $0.95/$4 per 1M — open-weights. Very low hallucination, HLE-with-tools 54.0. Roughly Opus-tier value at a fraction of the cost.

#1 open-weights model (AA Index)

GLM-5.2

Leading open-weights model on the AA Intelligence Index (51). MIT, 1M context, SWE-Pro 62.1, 28% hallucination vs 86% for GPT-5.5.

New Grok — built for coding / agentic

Grok 4.5

GA 8 Jul, trained on real Cursor sessions. AA Index 54 (4th globally), Terminal-Bench 83.3, $2/$6 — ~60% cheaper than Opus 4.8.

Cheapest viable at scale

DeepSeek V4-Flash

$0.14/$0.28 per 1M, SWE-V 79%, 1M context. For translation, tagging, classification and high-volume calls.

Longest autonomous tasks

Qwen 3.7-Max

Architected for ~35-hour autonomous runs and 1,000+ tool calls. AA Index 56.6, SWE-Pro 60.6.

Best balanced Claude for docs / decks

Claude Sonnet 5

Near-Opus agentic quality at $3/$15 (intro $2/$10 to Aug 31). SWE-Pro 63.2 > GPT-5.5.

Best self-hosted / data privacy

GLM-5.2 / Kimi K2.6 / DeepSeek

Open weights — skip API cost, keep data on your own cluster. GLM & Kimi lead capability; DeepSeek leads price.

Voice (Egyptian Arabic)

Live Egyptian conversation (S2S)

gpt-realtime-1.5 / Grok Voice Agent

Both strong on Egyptian. Grok at $0.05/min vs OpenAI ~$0.18–0.46/min — Grok far cheaper & lower-latency; OpenAI edges naturalness.

Best Egyptian TTS

ElevenLabs v3

Only vendor explicitly confirming Egyptian (plus Saudi/Levantine/Gulf/Maghrebi). Best naturalness, emotion & cloning.

Cheapest STT for Arabic

Grok STT

$0.10/hr batch, $0.20/hr streaming. 25+ languages, speaker diarization and word-level timestamps.

Image generation

Best overall (English)

GPT Image 2

Arena Elo 1339, 99% text accuracy, reasoning-powered. $0.053 (med) / $0.211 (high) per image.

Best Arabic / Egyptian

GPT Image 2

Only premium model with pixel-perfect Arabic confirmed. Qwen-Image 2.0 and HunyuanImage 3.0 are strong multilingual alternatives.

Best open + Arabic (self-host)

HunyuanImage 3.0

80B MoE, multilingual incl. Arabic, open OpenMDW licence — free to self-host on your own GPUs.

Sources & methodology

Primary benchmark leaderboards (verified 6 Jul 2026)

Artificial Analysis Intelligence Index (v4.1) — Fable 5 60, GPT-5.6 Sol 59, Qwen 3.7-Max 56.6, Grok 4.5 54, Sonnet 5 53, GLM-5.2 51 (#1 open), DeepSeek V4-Pro 50.4. source →
Vals.ai SWE-Bench Verified (updated 1 Jul) — Fable 5 95.0, Opus 4.8 88.6, Gemini 3.5 Flash 78.8. source →
Morphllm SWE-Bench Pro — Fable 5 80.3, Opus 4.8 69.2, Sonnet 5 63.2, GLM-5.2 62.1, Qwen 3.7-Max 60.6, Kimi K2.6 58.6. source →
AA-Omniscience (hallucination) — GLM-5.2 28% hallucination vs GPT-5.5 86%. source →
AA Image Arena (text-to-image) — GPT Image 2 1339, Reve 2.0 1281, MAI-Image-2.5 1271, Cosmos3 1227 (top open-weight). source →

Model & pricing sources

Anthropic Claude catalog — Fable 5 $10/$50, Opus 4.8 $5/$25, Sonnet 5 $3/$15 (intro $2/$10 to 31 Aug). All 1M context. source →
Google Gemini — 3.5 Flash $1.50/$9 is the current LLM entry; 3.5 Pro delayed (~17 Jul). source →
Moonshot Kimi K2.6 & K2.7-Code $0.95/$4 (cache-hit $0.19), 256K, Modified MIT. source →
DeepSeek V4-Pro $0.44/$0.87, V4-Flash $0.14/$0.28, 1M context. source →

Methodology

Benchmark-anchored scoring: the score is a 0–100 composite anchored to SWE-Bench Verified/Pro, HLE, GPQA, AA Intelligence Index and AA-Omniscience. Displayed sub-scores are analytical mappings from those benchmarks — not vendor marketing claims. OpenAI & xAI report agentic-coding indices & Terminal-Bench rather than SWE-Bench Pro.
Confidence flags on every row: ✅ verified from a primary/neutral-aggregator source; ⚠️ partial or estimated; ❓ released but no public benchmarks yet.
Egyptian Arabic gate (voice): retained. Only ElevenLabs v3 confirms Egyptian; GLM-ASR-2512 confirms generic Arabic. GLM-4-Voice (CN/EN only) fails the gate.
Still not scored (benchmarks pending): Google Gemini 3.5 Pro (delayed, ~17 Jul, enterprise preview). No model newer than GPT-5.6 / Grok 4.5 / GLM-5.2 confirmed as of 6 Jul 2026.

AI Markets

AI leaders and their capital cycle — capex commitments alongside the month’s market moves.

AI complex — share performance, year to date

Nvidia (NVDA)

$218.29

-4.45% today

Microsoft (MSFT)

$495.63

-2.84% today

Alphabet (GOOGL)

$338.50

-1.16% today

Meta (META)

$648.03

+6.12% today

Chips

Jensen Huang explains why Nvidia will grow an astounding 70% next year

(10 Sep) Nvidia CEO Jensen Huang projected another year of significant growth for the company, attributing the forecast to broad industry demand while denying that its recent deals are circular.

Read full article → TechCrunch

Policy

AI regulation calls grow in DC after researcher's extinction warning

(11 Sep) Members of the U.S. Congress are pushing for new AI regulations following warnings from an Anthropic researcher about the potential risks of rapid AI self-improvement.

Read full article → CNBC

Industry

Anthropic details distillation campaigns from Alibaba, Moonshot AI, and DeepSeek

(10 Sep) Anthropic published a report alleging persistent and escalating unauthorized distillation attacks by several China-based AI companies targeting its Claude models.

Read full article → TechCrunch

Industry

Amazon gives OpenAI's ad business a boost, letting its advertisers into ChatGPT

(10 Sep) Amazon is allowing its advertisers to place campaigns within ChatGPT, marking a shift in the company's previous reluctance to open its webstore to external AI platforms.

Read full article → CNBC

Infrastructure & Capital

Funding

Mecka AI nears $500M valuation in Sequoia-led deal amid rush for robot training data · $500M

(11 Sep) Robotics data startup Mecka AI is finalizing a new funding round led by Sequoia, reaching a valuation of $500 million as demand for robot training data accelerates.

Read full article → TechCrunch

Funding

Maven Robotics wants to steal your robot deployment deal · $100M

(10 Sep) Maven Robotics emerged from stealth with active deployments and a $100 million Series A funding round to expand its robot deployment services.

Read full article → TechCrunch

Funding

AI research startup Listen Labs scrubbed a $1.5B funding round for Salesforce talks · $1.5B

(10 Sep) AI research startup Listen Labs abandoned a signed $1.5 billion Series C term sheet from Menlo Ventures to pursue strategic discussions with Salesforce.

Read full article → TechCrunch