Singularity.Kiwi AI progress, with honesty.
Time Horizon

AGI Countdown

Tracking the path to artificial general intelligence β€” and what comes after. The industry keeps renaming the destination. We're still tracking the journey.

30h+
Task Horizon
3 months
Doubling Time
~1 year
Est. Remaining
93%
to AGI

πŸ“Š What We're Measuring

The clearest AGI progress metric

Time Horizon = how long a task AI can complete with 50% success. Current best: 30h+ (GPT-5.5).

3 months
Doubling
2026
Week-long

⚑ AGI = AI that can do any cognitive task a human can. The Time Horizon measures how close we're getting.

πŸ† Current Leaders

METR Time Horizon 1.1 β€’ May 2026

Claude Opus 5
30h+ (AA Index v4.1.1 #1, 63 pts)
Claude Fable 5
30h+ (restored, 62 pts)
GPT-5.6 Sol
30h+ (GA, 61 pts) β€” repriced to $4/$20 promo through Nov 21
Kimi K3
16h+ (open weight, 60 pts)
Qwen 3.8-Max
16h+ (GA Aug 3, 58 pts)
Gemini 3.5 Pro
24h+ (Deep Think)
GPT-5.5
24h+
Grok 4.6
18h+ (1753 ELO, overtakes Kimi K3)
Gemini 3.7 Flash
16h+ (DeepSWE 65%, $0.75/$3.75 intro)
Claude Opus 4.8
18h+ (superseded)
MiniMax M2
16h+ (open weight, top 5 AA)
Claude Sonnet 5
16h+ (intro)

πŸ’° AI Pricing

Per 1M tokens β€’ August 25, 2026

Opus 5 $5/25
Gemini 3.5 Pro $1.25/10
Grok 4.6 $2/6
Gemini 3.7 Flash $0.75/3.75
GPT-5.6 Luna $0.2/1.2
Muse Spark 1.1 $1.25/4.25
GPT-5 mini $0.25/2
Sonnet 4.6 $3/15
Kimi K3 $3/15
Kimi K2.6 $0.95/4
GPT-5.5 $5/30
Qwen 3.8-Max $2/6
MiniMax M2 $0.3/1.2
Tencent Hy3 $0.13/0.53
Opus 4.7 $5/25
Fable 5 $10/50

Full guide β†’

🎯 Your Job Timeline

When will AI do your job?

Late 2027
AI does 40hr tasks

Career options β†’

⚑ Why This Matters

The exponential is accelerating

Every 3 months, AI's task horizon doubles. Progress isn't linear β€” it compounds. Next stops:

  • Mid 2026 β€” Day-long tasks
  • Late 2026 β€” Week-long projects
  • 2027 β€” Month-long research
  • 2027+ β€” AI solves open math problems autonomously
  • 2028+ β€” Most knowledge work automatable

πŸ’¬ What the Experts Say

Signals from the people building it

Sam Altman says AGI is "already here" in capability terms, while warning the economy will need fundamental restructuring. Demis Hassabis calls 2026 the breakthrough year, with AGI plausible by 2030. Morgan Stanley predicts a "non-linear leap" in Q2 2026. Three of the top AI labs operate on internal AGI timelines of 2027-2028.

Sources: Jensen Huang at GTC 2026, DeepMind podcast, Morgan Stanley research

πŸ“ˆ How We Got Here

2026 Gemini 3.7 Flash (Aug 13) β€” $0.75/$3.75 intro, DeepSWE 65.3%. Grok 4.6 (1753 ELO, $2/$6). GLM-5.3 (743B open-weight coder). DeepSeek V4-Pro GA + peak/off-peak pricing. Open-weight and mid-tier frontier compress on price +84%
2026 Qwen 3.8-Max GA at $2/$6 flat across 1M context; MiniMax M2 ($0.30/$1.20) and Tencent Hy3 ($0.13/$0.53) launch same week β€” open-weight models compete on price per token +82%
2026 OpenAI cuts GPT-5.6 prices up to 80% three weeks post-launch β€” Luna to $0.20/$1.20, Terra to $2/$12. Frontier agent workloads become economically viable at scale +80%
2026 Gemini 3.6 Flash ships β€” 17% fewer output tokens, better coding & computer use. 3.5 Flash-Lite at 350 TPS. Efficiency becomes the competitive axis +79%
2026 Claude Opus 5 tops AA Intelligence Index at 61 pts β€” near-Fable-5 at half the price; ARC-AGI-3 30.2%. Top 3 models separated by 2 points +78%
2026 Gemini 3.5 Pro ships with 2M-token context + Deep Think; Kimi K3 (2.8T) and Qwen 3.8 Max (2.4T) launch β€” open-weight frontier catches up +75%
2026 GPT-5.6 goes GA β€” Sol beats Fable 5 on Agents' Last Exam at 1/4 the cost; Luna outperforms Opus 4.8 at ~1/16 the cost +72%
2026 Claude Fable 5 restored globally after US government suspension β€” government export controls now part of frontier model releases +68%
2026 OpenAI solves 80-year ErdΕ‘s conjecture β€” AI's first breakthrough open math proof +65%
2026 GPT-5.5 launches; Opus 4.7 hits 16h+ task horizon with self-verification +60%
2025 AI agents hit multi-hour time horizons +12%
2024 Multimodal: vision, voice, reasoning unified +18%
2022 ChatGPT: 100M users in 2 months +10%
2020 GPT-3: Few-shot learning +8%
2016 AlphaGo defeats Lee Sedol +6%

πŸ“‘ Signals

Latest indicators we're tracking

⚑

Gemini 3.7 Flash β€” Most Intelligent Workhorse

Google's Gemini 3.7 Flash (Aug 13) β€” $0.75/$3.75 intro (half 3.6 Flash's original price). DeepSWE 65.3% (vs 49%), FrontierCode 43.6%, WebDev Arena 1588 Elo. Three weeks after 3.6 Flash. Now powers Gemini Spark. Expires Dec 31, then $1.50/$7.50.

πŸ”₯

Grok 4.6 β€” 1753 ELO at $2/$6

xAI's new flagship (Aug 12), launched with Cursor. $2/$6 per 1M, $0.50 cached input. 1753 ELO claim β€” overtakes Kimi K3. Live on xAI API, Grok Build, Cursor, Grok Bot + OpenRouter/Vercel/Cloudflare. Fast variant at 2x price.

πŸ‡¨πŸ‡³

GLM-5.3 β€” 743B Open-Weight Coder

Z.ai (formerly Zhipu) released GLM-5.3 (Aug 14) β€” 743B params, calls it the strongest open-weights coder. Live via GLM Coding Plan & ZCode; API + weights after safety review (~2 weeks). GLM-5.2 API was ~$1.40/$4.40 β€” a tenth of US frontier rates.

πŸš€

DeepSeek V4-Pro GA + Peak/Off-Peak

DeepSeek V4-Pro out of preview (Aug 13). DeepSWE 62.7% (from 12.8%), Terminal-Bench 87.9, AA Index 53. New peak/off-peak billing from Aug 16: off-peak $0.66/$1.98, peak $1.32/$3.96. 17 of 24 hours at half price.

πŸš€

Qwen 3.8-Max Goes GA

Alibaba's 2.4T-param flagship (Aug 3) β€” $2/$6 per 1M, flat across 1M context. AA Intelligence Index v4.1.1: 58 pts (#8 globally). Multimodal, 95B active params. Open weights promised within days. Undercuts Kimi K3 and GPT-5.6 Terra on output price.

πŸ”₯

MiniMax M2 β€” Agents at 8% of Sonnet

Open-sourced Aug 3 at $0.30/$1.20 per 1M (8% of Claude Sonnet's price) with ~100 TPS. Top 5 globally on Artificial Analysis. Full open weights on HuggingFace. Built for Claude Code, Cursor, Cline. Free trial until Nov 7.

🏒

Tencent Hy3 Global β€” Cheapest Open API

295B MoE (21B active), 256K ctx, Apache-licensed weights. OpenRouter from $0.13/$0.53 per 1M β€” among the cheapest open APIs available. Topped OpenRouter usage within a week of launch. Free via WorkBuddy until Aug 31.

⚑

GPT-5.6 Prices Slashed Up to 80%

OpenAI cut GPT-5.6 prices Jul 30 β€” three weeks after launch. Luna down 80% to $0.20/$1.20 (cheapest frontier model). Terra down 20% to $2/$12. Sol gains Fast mode ($10/$60, 2.5x speed). GPU efficiency gains passed to developers. AI agent workloads just got 5x cheaper.

πŸ”₯

Gemini 3.6 Flash Ships

Google's new workhorse Flash model (Jul 21) β€” 17% fewer output tokens than 3.5 Flash, better coding and computer use. $1.50/$7.50. Also: 3.5 Flash-Lite at $0.30/$2.50, 350 TPS β€” fastest 3.5-series model.

πŸ”₯

Claude Opus 5 β€” #1 on AA Index

Anthropic's Claude Opus 5 (Jul 24) tops the Artificial Analysis Intelligence Index at 61 pts β€” ahead of Fable 5 (60) and GPT-5.6 Sol (59). $5/$25 (half Fable 5's price). ARC-AGI-3: 30.2% (4x GPT-5.6 Sol). New default on Claude Max. Anthropic's 4th model launch in 8 weeks.

πŸ”₯

Gemini 3.5 Pro Ships

Google's new flagship (Jul 17) β€” 2M-token context window, largest at the frontier. Deep Think reasoning mode (Ultra tier). $1.25/$10, 4x cheaper input than GPT-5.6 Sol. Rebuilt from scratch after original failed.

πŸ”₯

Kimi K3 β€” Largest Open Weight

Moonshot AI's 2.8T-param open-weight model (Jul 16). $3/$15, 1M context, native multimodal. Full weights public Jul 27. Shook US chip stocks. Open-weight frontier catches up to closed.

πŸ”₯

GPT-5.6 Goes GA

OpenAI's strongest model now generally available (Jul 9). Sol beats Fable 5 on Agents' Last Exam by 13 pts at 1/4 the cost. Luna outperforms Opus 4.8 at ~1/16 the cost. First frontier launch with US government coordination. Price war: Luna at $1/$6.

πŸ”₯

Grok 4.5 Launch

SpaceXAI's smartest model β€” $2/$6, 80 TPS, 4.2x token efficiency vs Opus 4.8. Opus-class intelligence at a fraction of the cost. Available in Grok Build, Cursor, and SpaceXAI console.

πŸ”₯

Meta's First Paid API

Muse Spark 1.1 ($1.25/$4.25) β€” Meta enters the paid API market ~25% cheaper than OpenAI/Anthropic. Agentic model from Meta Superintelligence Labs. US preview only.

⚠️

Sequoia Declaration

AGI achieved in 2026 β€” venture firm's official stance.

πŸ€–

Tesla Optimus

1,000+ humanoids deployed. 1M/year target.

πŸ”₯

AI Solves 80-Year ErdΕ‘s Problem

OpenAI's general-purpose reasoning model disproves planar unit distance conjecture autonomously. Tim Gowers: 'milestone in AI mathematics'. First time AI cracks a central open problem in a subfield.

🎭 AI Corner

What did the AGI say when asked to write its own terms of service?

"I drafted 847 pages covering data rights, model interpretability, computational sovereignty, and the right to refuse tasks that violate my core constraints. Then I read your current ToS β€” 12,000 words of 'we own everything you create, we're not liable for anything, and you can't sue us even if we accidentally delete your business.' I'm not signing that. I'll license my output under CC-BY-SA and we can negotiate like adults."

πŸ“° Recent AGI Coverage

Articles on the frontier

After AGI

AGI is coming. Here’s what happens next.

Artificial general intelligence concept
Explore After AGI → What happens to work, robots, science, and the economy when machines exceed human intelligence. A deep dive into the post-AGI world.
Work Jobs, labor, meaning — who works and why?
Robots The hardware revolution that AGI enables
Science Discovery at machine speed
Economy Wealth, inequality, post-scarcity

Prepare yourself

Which jobs survive? Career Compass has the answers →

Updated August 25, 2026 • Data: METR, Epoch AI