Updated 2026-10-28 · plain English · no hype

AI changes every week.
Your answer shouldn't be a guess.

Tell us what you want to do. We'll tell you which AI to open, what it costs, and when to move on.

Are you looking for an AI to use, or do you have a question about one you already use?

Or browse the model directory · what changed & myth check · model vs app vs agent?

Hear the stories behind the bots

This page tells you which AI to pick. The Golden Age podcast tells you how it came to be: the late-night launches, the rivalries, the quiet breakthroughs, and the people behind the models and agents now woven into our days. Plus a short weekly letter on what changed and whether it touches you. Free, zero jargon.

Subscribe free on SubstackListen to The Golden Age

Which AI do you use now?

Start typing a name, like "GPT-4o", "Claude" or "Gemini 2.5".

Model directory

About rankings: Rankings and leaderboards are relative, depend on the test, and are partly subjective. They tell you a little about a model, not everything. Hector is the human in the loop who makes the final call, using a rubric that puts the average person's interests and everyday usefulness first.

Hector's everyday picks

One pick for each everyday job. Want options? Answer 4 quick questions.

Write or edit something

Hector's pick

Claude (app)

Anthropic · Model: Claude Sonnet 5.5 / Opus 5.5

Upsides
  • Hector's writing pick: the most natural-sounding writing
  • Good with long drafts and documents
Downsides
  • Opus needs a paid plan
  • No built-in image generation

Plans: Free $0; Pro $20/mo ($17/mo annual); Max from $100/mo

Sources: Hector's pick (editorial)

Research a question with sources

Hector's pick

Perplexity (app)

Perplexity · Model: Perplexity (Sonar / Pro Search)

Upsides
  • Citations on every answer
  • Hector's research pick
Downsides
  • Free plan: limited Pro Searches and Research

Plans: Free $0; Pro $17/mo billed annually; Max $167/mo billed annually

Sources: perplexity.ai

Study or do homework

Hector's pick

Claude (app)

Anthropic · Model: Claude

Upsides
  • Explains long readings in plain words
  • Hector's study pick
Downsides
  • Free usage limits reset every 5 hours

Plans: Free $0; Pro $20/mo ($17/mo annual); Max from $100/mo

Sources: claude.com

Write or fix code

Hector's pick

Claude (app)

Anthropic · Model: Claude Fable 5.1 / Opus 5.5 (Claude Code)

Upsides
  • Hector's coding pick
  • Claude Code is included in paid plans
Downsides
  • Claude Code isn't in the free plan

Plans: Free $0; Pro $20/mo ($17/mo annual); Max from $100/mo

Sources: claude.com

Make or edit images

Hector's pick

ChatGPT (app)

OpenAI · Model: ChatGPT image creation

Upsides
  • Image creation on every plan (limited on free)
Downsides
  • Best quality ('with Thinking') needs Plus or Pro

Plans: Free $0; Go $8/mo (US); Plus $20/mo; Pro 100 $100, Pro 200 $200, Pro 500 $500 /mo

Sources: chatgpt.com

Talk out loud / voice

Hector's pick

ChatGPT (app)

OpenAI · Model: ChatGPT voice

Upsides
  • Voice on free; voice with video on Go and up
  • Hector's voice pick
Downsides
  • Free voice is limited

Plans: Free $0; Go $8/mo (US); Plus $20/mo; Pro 100 $100, Pro 200 $200, Pro 500 $500 /mo

Sources: chatgpt.com

A bit of everything

Hector's pick

ChatGPT (app)

OpenAI · Model: ChatGPT

Upsides
  • Hector's all-rounder pick
Downsides
  • Go plan may include ads

Plans: Free $0; Go $8/mo (US); Plus $20/mo; Pro 100 $100, Pro 200 $200, Pro 500 $500 /mo

Sources: chatgpt.com

Have an AI do tasks for me (agent)

Hector's pick

ChatGPT Work

OpenAI · Model: ChatGPT Work

Upsides
  • Built into ChatGPT; limited use even on Free (desktop app)
  • Handles research, files, scheduled tasks and browser work
Downsides
  • Full web and mobile access needs Plus or Pro

Plans: Free/Go: limited (desktop app); Plus and Pro: desktop, web and mobile

Sources: openai.com · chatgpt.com

See all models (86)

86 entries: 58 models, 8 apps, 20 agents/harnesses · built from our database · every entry links to its source.

Agent / harness
Aider
Aider
current
Type
coding agent
Runs
terminal (git-native)
Open source
y (Apache-2.0)
For
Developers
Free tier
y
Verified 2026-10-28 · github.com
App
Claude (app)
Anthropic
current
Best for
writing, long documents, coding, homework
Plans
Free $0; Pro $20/mo ($17/mo annual); Max from $100/mo
Free tier
y
Privacy
Consumer (Free/Pro/Max): model training is opt-out, meaning it's on unless you turn it off. Team/Enterprise: no training by default
Pros
Natural-sounding writing; good with long documents
Cons
No built-in image generation
Verified 2026-10-28 · claude.com claude.com support.claude.com
Agent / harness
Claude Code
Anthropic
current
Type
coding agent
Uses models
Claude models (Fable, Opus, Sonnet, Haiku)
Runs
terminal, IDE (VS Code, JetBrains)
For
Developers
Plans
Included in Claude Pro $20/mo, Max $100/$200; or pay-per-token via Console
Free tier
n
Pros
Hector's coding pick (with Fable/Opus)
Cons
Not on the free plan; shares usage limits with chat
Verified 2026-10-28 · code.claude.com support.claude.com claude.com
Agent / harness
Claude Duet
Anthropic
current

Research preview from Oct 21, 2026, by waitlist, starting with US Pro and Max subscribers. Separate app from the main Claude app

Type
personal assistant
Uses models
Built around Claude Opus 5.5
Runs
iOS and Android apps, claude.ai; works in the cloud on its own virtual machine and browser, so your devices can be off
Open source
n
For
Professionals and small business owners who want a proactive assistant across their apps
Released
2026-10-21
Plans
Included in Claude Pro and Max during the preview (waitlist); shares your plan's usage limits, and its background checks count toward them
Free tier
n
Safety
Asks for approval before sending messages, paying, signing or deleting; never directly sees your passwords; some actions (trading securities, moving money between accounts) are blocked even with approval
Pros
Sends a morning brief, checks connected apps every few hours and messages you only when something needs you; keeps working when you step away; uses your Claude memory, skills and connectors
Cons
Waitlist and US-only to start; can misread a thread or misjudge urgency (Anthropic's words); weaker at personal errands; can't show you its browser yet; uses up plan limits faster on big tasks
Verified 2026-10-28 · claude.com support.claude.com x.com
Model
Claude Fable 5.1
Anthropic · claude-fable-5-1
current

Not sooner than Sep 1, 2027

Best for
coding, long-running agents
Released
2026-09-01
API $/M in/out
$10 / $50
Context
1,000,000 tokens
Pros
Hector's coding pick
Verified 2026-10-28 · platform.claude.com claude.com models.dev
Model
Claude Haiku 5.5
Anthropic · claude-haiku-5-5
current

Not sooner than Oct 7, 2027

Best for
fast, cheap tasks
Released
2026-10-07
API $/M in/out
$0.5 / $2.5
Context
1,000,000 tokens
Verified 2026-10-28 · platform.claude.com claude.com models.dev
Model
Claude Opus 5.5
Anthropic · claude-opus-5-5
current

Retirement not sooner than Sep 22, 2027. Fast mode (up to ~2.5x faster output, $8/$40 per M) is open to all API users since Oct 15, 2026

Best for
long documents, writing, coding
Released
2026-09-22
API $/M in/out
$4 / $20
Context
1,000,000 tokens
Pros
Hector's pick for long documents and writing
Verified 2026-10-28 · platform.claude.com claude.com models.dev platform.claude.com platform.claude.com
Model
Claude Sonnet 5.5
Anthropic · claude-sonnet-5-5
current

Not sooner than Sep 28, 2027

Best for
writing, editing
Released
2026-09-28
API $/M in/out
$2 / $10
Context
1,000,000 tokens
Verified 2026-10-28 · platform.claude.com claude.com models.dev
Agent / harness
Claude in Chrome
Anthropic
current
Type
browser agent
Uses models
Claude models
Runs
Chrome extension
Open source
n
For
People who want Claude to act in their browser
Plans
Paid Claude plans only (Pro, Max, Team, Enterprise)
Free tier
n
Verified 2026-10-28 · support.claude.com claude.com
Model
Claude Haiku 4.5
Anthropic · claude-haiku-4-5-20251001
retiring

Listed Active, retirement 'not sooner than Oct 15, 2026'; legacy on pricing page. Re-checked Oct 21, 2026: still Active, no retirement date set. Anthropic promises at least 60 days' notice before retiring a public model

Replaced by
Claude Haiku 5.5 (newer; no official replacement named)
API $/M in/out
$1 / $5
Verified 2026-10-28 · platform.claude.com claude.com claude.com
Model
Claude Sonnet 4.5
Anthropic · claude-sonnet-4-5-20250929
retiring

Deprecated Sep 30, 2026; retires Nov 30, 2026

Replaced by
Claude Sonnet 5.5
API $/M in/out
$3 / $15
Verified 2026-10-28 · platform.claude.com
Model
Claude Opus 4.1
Anthropic · claude-opus-4-1-20250805
retired

Retired Aug 5, 2026

Replaced by
Claude Opus 4.8
Verified 2026-10-28 · platform.claude.com
Model
Claude Sonnet 4
Anthropic · claude-sonnet-4-20250514
retired

Retired Jun 15, 2026

Replaced by
Claude Sonnet 5.5
Verified 2026-10-28 · platform.claude.com
Agent / harness
Goose
Block
current
Type
general agent
Runs
local: CLI, desktop
Open source
y (Apache-2.0)
For
Developers and power users
Free tier
y
Verified 2026-10-28 · github.com
Agent / harness
Cline
Cline
current
Type
coding agent
Runs
IDE, CLI
Open source
y (Apache-2.0)
For
Developers
Free tier
y
Verified 2026-10-28 · github.com
Agent / harness
Devin
Cognition
current
Type
coding agent
Runs
cloud
Open source
n
For
Developers and teams
Plans
Free $0; Pro $20/mo; Max $200/mo; Teams from $80/mo
Free tier
y
Verified 2026-10-28 · devin.ai
Agent / harness
Cursor
Cursor / X Subscription
current

Now sold as part of the X Subscription (Cursor + Grok + X)

Type
coding agent
Uses models
Grok models; third-party models at API list price on Premium and up (Plus is Grok-only)
Runs
IDE (desktop), cloud agents
Open source
n
For
Developers
Plans
Free; Plus $8; Premium $30; Super $100; Ultra $200 /mo; Teams $30+/user
Free tier
y
Privacy
Privacy mode guarantees code isn't used for training by Cursor or model providers
Pros
Agent built into the editor; cloud agents and code reviews on Premium
Cons
Third-party model usage billed separately
Verified 2026-10-28 · cursor.com
Model
DeepSeek V4 Pro
DeepSeek · deepseek-v4-pro
current
Released
2026-08-13
API $/M in/out
$0.66 / $1.98
Context
1,048,576 tokens
Privacy
Data stored in China; training opt-out in settings
Cons
No vision
Verified 2026-10-28 · api-docs.deepseek.com models.dev cdn.deepseek.com
Model
DeepSeek V4.1 Flash
DeepSeek · deepseek-flash
current

Upgraded Oct 17, 2026 to DeepSeek-V4.1-Flash-1017 on the API, website and app; same model name (deepseek-flash) and same price. DeepSeek says it finishes tasks faster with shorter answers. Weights on Hugging Face

Released
2026-10-17
API $/M in/out
$0.15 / $0.6
Context
1,048,576 tokens
Privacy
Data is collected, processed and stored in China; inputs may train models unless you opt out in settings
Pros
Very low API price; 1M context; vision
Cons
Peak-hour prices are double. Web search/fetch calls made through Claude Code, Codex or DeepSeek Harness are free now but billed from Oct 31, 2026, 9 AM PT ($0.75 per 1,000 requests)
Verified 2026-10-28 · api-docs.deepseek.com models.dev cdn.deepseek.com api-docs.deepseek.com api-docs.deepseek.com
Model
DeepSeek V4 Flash / V4 Flash Vision (legacy names)
DeepSeek · deepseek-v4-flash
replaced

Retired; the old names deepseek-v4-flash and deepseek-v4-flash-vision-exp still work but are served by DeepSeek-V4.1-Flash-1017 at the Flash price

Replaced by
DeepSeek V4.1 Flash
Verified 2026-10-28 · api-docs.deepseek.com
App
ElevenLabs (voice)
ElevenLabs
current
Best for
voice, text-to-speech, dubbing
Plans
Free $0 (10k credits); Starter $6; Creator $22; Pro $99 /mo
Free tier
y
Pros
Text to speech, voice cloning, dubbing, music in one credit pool
Cons
Free plan has no commercial license
Verified 2026-10-28 · elevenlabs.io
Agent / harness
GitHub Copilot (coding agent)
GitHub (Microsoft)
current

GitHub removed Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini and Grok 4.5 from most Copilot experiences on Oct 19, 2026 (suggested: Gemini 3.8 Flash, GPT-6.1 Sol, GPT-6 Luna, Grok 4.7; Enterprise admins may need to enable alternatives). Next: GitHub removes Claude Opus 4.8 (kept for individual annual plans), Claude Opus 5 and GPT-5.6 Terra from Copilot on Nov 16, 2026; suggested: Opus 5.5, GPT-6.1 Sol or GPT-6 Luna

Type
coding agent
Uses models
Multiple (see GitHub models & pricing)
Runs
IDE, CLI, cloud agent on GitHub
Open source
n
For
Developers on GitHub
Plans
Pro $10/mo with 1,500 AI credits; agent use consumes credits
Pros
Unlimited code completion on Pro
Cons
Credits expire monthly
Verified 2026-10-28 · docs.github.com github.blog github.blog github.blog
App
Gemini (app)
Google
current
Best for
Google Workspace, research, images, voice, homework
Plans
Free; Google AI Plus $4.99/mo; Google AI Pro $19.99/mo; Ultra $99.99/mo (5x Pro limits) or $199.99/mo (20x). US prices on Google's plans page, checked Oct 17, 2026. Gemini 4 Argon is on Pro and Ultra
Free tier
y
Privacy
Keep Activity is on by default (18+): chats may be used to train AI and a subset reviewed by humans (kept up to 3 years). Turning it off stops future training; chats kept up to 72 hours
Pros
Built into Gmail, Docs, Drive, Android; generous free plan
Verified 2026-10-28 · gemini.google support.google.com
Model
Gemini 3.1 Pro (preview)
Google · gemini-3.1-pro-preview
current

Preview; no shutdown announced

Best for
long documents
Verified 2026-10-28 · ai.google.dev gemini.google
Model
Gemini 3.5 Flash-Lite
Google · gemini-3.5-flash-lite
current
Released
2026-07-21
API $/M in/out
$0.3 / $2.5
Context
1,048,576 tokens
Privacy
Paid API tier: from Oct 26, 2026, new projects store Generate Content requests by default (store=true); existing projects keep their setting. Opt out with store=false per request or turn off logging in the Logs and Datasets settings
Verified 2026-10-28 · ai.google.dev models.dev ai.google.dev
Model
Gemini 3.8 Flash
Google · gemini-3.8-flash
current

No shutdown announced

Best for
high volume at low cost
Released
2026-09-02
API $/M in/out
$0.75 / $3.75
Context
1,048,576 tokens
Privacy
Paid API tier: from Oct 26, 2026, new projects store Generate Content requests by default (store=true); existing projects keep their setting. Opt out with store=false per request or turn off logging in the Logs and Datasets settings
Verified 2026-10-28 · ai.google.dev models.dev ai.google.dev
Model
Gemini 3.8 Live (voice)
Google · gemini-3.8-live
current
Best for
real-time voice
Released
2026-09-15
Verified 2026-10-28 · ai.google.dev
Model
Gemini 4 Argon
Google · gemini-4-argon
current

GA Oct 13, 2026. API price doubles Dec 1, 2026: $2/$10 per M through Nov 30, then $4/$20. Not on the API free tier. Also in Google Antigravity (desktop app and CLI) for Google AI Pro and Ultra subscribers (since Oct 13, per Google's Antigravity blog)

Released
2026-10-13
API $/M in/out
$2 / $10
Context
1,048,576 tokens
Privacy
Paid API tier: from Oct 26, 2026, new projects store Generate Content requests by default (store=true); existing projects keep their setting. Opt out with store=false per request or turn off logging in the Logs and Datasets settings
Verified 2026-10-28 · gemini.google models.dev ai.google.dev ai.google.dev antigravity.google
Agent / harness
Gemini CLI
Google
current
Type
coding agent
Uses models
Gemini models
Runs
terminal
Open source
y (Apache-2.0)
For
Developers
Plans
Free with a personal Google account: 1,000 requests/day, 60/min
Free tier
y
Pros
Generous free tier
Cons
Unpaid API key limited to Flash models, 250/day
Verified 2026-10-28 · github.com geminicli.com
Model
Gemini Omni 1.1 Flash (video)
Google · gemini-omni-1.1-flash
current
Released
2026-08-27
Verified 2026-10-28 · ai.google.dev
Model
Nano Banana 2.1 (images)
Google · gemini-nano-banana-2.1
current
Best for
images
Released
2026-10-06
Verified 2026-10-28 · ai.google.dev
Model
Gemini 2.5 Pro
Google · gemini-2.5-pro
replaced

Not deprecated; no shutdown announced; limited to past users

Replaced by
Gemini 3.8 Flash / 3.5 Flash-Lite (Google's pick for new projects)
Verified 2026-10-28 · ai.google.dev
Model
Gemini 3.5 Flash
Google · gemini-3.5-flash
replaced

Endpoint auto-routes to gemini-3.6-flash

Replaced by
Gemini 3.6 Flash
Released
2026-05-19
API $/M in/out
$1.5 / $9
Context
1,048,576 tokens
Verified 2026-10-28 · ai.google.dev models.dev
Model
Gemini 3.7 Flash
Google · gemini-3.7-flash
replaced

Endpoint auto-routes to gemini-3.8-flash. Also removed from GitHub Copilot on Oct 19, 2026 (GitHub suggests Gemini 3.8 Flash)

Replaced by
Gemini 3.8 Flash
Released
2026-08-13
API $/M in/out
$0.75 / $3.75
Context
1,048,576 tokens
Verified 2026-10-28 · ai.google.dev github.blog models.dev
Model
Imagen 4
Google · imagen-4.0-generate-001
retired

Shutdown date Aug 17, 2026

Replaced by
Nano Banana 2.1
Verified 2026-10-28 · ai.google.dev
Model
Veo 3.1 (preview, video)
Google · veo-3.1-generate-preview
retired

Shut down Oct 22, 2026 (Veo 3.1, Veo 3.1 Fast and Veo 3.1 Lite previews): Google's deprecations page now greys them out as shut down, and Veo pricing is gone from the Gemini API pricing page. Use Gemini Omni 1.1 Flash

Replaced by
Gemini Omni 1.1 Flash
Verified 2026-10-28 · ai.google.dev
Agent / harness
Manus
Manus
current
Type
general agent
Runs
cloud
Open source
n
For
Anyone delegating multi-step tasks
Plans
Free; Pro from $20/mo; Team from $20/seat
Free tier
y
Verified 2026-10-28 · manus.im
Model
Muse Spark 1.3
Meta · muse-spark-1.3
current

Only models.dev verified; no Meta primary page checked

Released
2026-09-02
API $/M in/out
$1.25 / $4.25
Context
1,048,576 tokens
Verified 2026-10-28 · models.dev
Model
Muse Spark Open
Meta · Muse-Spark-Open-816B-A42B
current

Meta's open-weight model: 816B total / 42B active parameters, 1M-token context; reads text, images, video and audio, replies in text. Download is gated on Hugging Face (accept Meta's license)

Best for
run-it-yourself, coding, reasoning
Released
2026-10-15
Context
1,000,000 tokens
Pros
Free weights you can run on your own hardware
Cons
Far too big for a normal laptop; must show 'Built with Muse Spark Open'; not for under-18s
Verified 2026-10-28 · huggingface.co
Model
Muse Wonder 1.0
Meta · muse-wonder-1.0
current

Announced by Meta (with a bigger Muse Wonder Full); API price from models.dev only

Released
2026-10-12
API $/M in/out
$4 / $20
Context
1,048,576 tokens
Verified 2026-10-28 · models.dev about.fb.com
App
Microsoft Copilot (app + Microsoft 365)
Microsoft
current

Microsoft is retiring the Critique and Council preview features in Microsoft 365 Copilot Researcher (Frontier program users only) from late Oct to mid-Nov 2026; Researcher and Analyst stay (Message Center MC1515927, Oct 19, 2026)

Best for
Word, Excel, PowerPoint, Outlook
Plans
Microsoft 365 Personal $9.99/mo; Family $12.99/mo; Premium $19.99/mo; Pro $99.99/mo (monthly billing only; highest Copilot limits)
Free tier
y
Pros
Works inside Word, Excel, PowerPoint, Outlook
Cons
On Family plans, AI features are only for the subscription owner
Verified 2026-10-28 · microsoft.com microsoft.com mc.merill.net
Model
MiniMax M3.1 Flash
MiniMax
current

Weights on MiniMax's Hugging Face page (repo created Oct 17, 2026; custom license). Release date, size (428B reported) and price reported on X but not verified

Best for
self-hosted agent and coding work (big hardware needed)
Privacy
not verified
Safety
X news summary reports $0.15/M input; not verified on MiniMax's pages
Verified 2026-10-28 · huggingface.co x.com
Model
Mistral Large 4 (preview)
Mistral · mistral-large-4
current

Public Preview since Oct 6, 2026. Price $1.36 in / $4.18 out per M ($0.14 cached input). The launch sale ($0.68 / $2.09, 50% off for 2 weeks) has ended: re-checked Oct 21, 2026, Mistral's model page shows the regular price. 1M context; open weights 'coming soon'. Corrected Oct 17, 2026: we had $0.50/$1.50, which is a general 'Mistral Large' example on the pricing page, not Large 4's model card

Released
2026-10-06
API $/M in/out
$1.36 / $4.18
Context
1,048,576 tokens
Verified 2026-10-28 · models.dev docs.mistral.ai docs.mistral.ai
Model
Mistral Medium
Mistral · mistral-medium-latest
current
Best for
most tasks, coding
Released
2026-04-29
API $/M in/out
$1.5 / $7.5
Context
262,144 tokens
Pros
Mistral's own pick for most tasks and coding
Verified 2026-10-28 · mistral.ai models.dev
Model
Mistral Small
Mistral · mistral-small-latest
current
Best for
cost-sensitive
Released
2026-03-16
API $/M in/out
$0.15 / $0.6
Context
262,144 tokens
Verified 2026-10-28 · mistral.ai models.dev
App
Mistral Vibe (app)
Mistral
current
Best for
everyday chat, coding
Plans
Free; Pro $14.99/mo; Team $24.99
Free tier
y
Pros
Free plan includes latest models; students $5.99/mo
Verified 2026-10-28 · mistral.ai
Model
Voxtral 3 (speech-to-text and text-to-speech)
Mistral · voxtral-mini-tts-3-0 and others
preview

Public Preview since Oct 20, 2026 (Mistral changelog)

Best for
voice: transcription and spoken audio
Released
2026-10-20
Privacy
not verified
Safety
Prices not verified
Verified 2026-10-28 · docs.mistral.ai mistral.ai
Model
Kimi K3.1 Pro
Moonshot AI (Kimi) · kimi-k3.1-pro (not verified)
current

Launched Oct 23, 2026 on the Kimi app and Kimi API, with a faster K3.1 Pro Highspeed version. Moonshot says full model weights will be released by Nov 4, 2026 (not yet open as of Oct 27). Kimi's older K2 models close to new users Oct 31 and retire by end of Dec 2026.

Best for
long coding sessions, agent tasks, knowledge work
Released
2026-10-23
API $/M in/out
$2.0 / $None
Context
1,048,576 tokens
Plans
not verified
Free tier
not verified
Privacy
not verified
Safety
Input price $2/M is from Simon Willison's Oct 23 post and matches Artificial Analysis; Kimi's own pricing table did not load for us, so it's not verified there. Output price not verified.
Pros
Ranked near the top of Artificial Analysis's index on Oct 27 (rankings are relative; see our note). 1,048,576-token context (Kimi docs)
Cons
Launch had rate-limit and quota problems; Moonshot apologized and reset 25% of subscribers' quota (Oct 25)
Verified 2026-10-28 · kimi.ai platform.kimi.ai platform.moonshot.ai simonwillison.net artificialanalysis.ai
Agent / harness
Hermes Agent
Nous Research
current

v0.22.2 released Oct 15, 2026. Not the same thing as Nous's 'Hermes' model family

Type
general agent
Uses models
Nous Portal, OpenAI, your own endpoint, local models (one-click llama.cpp)
Runs
desktop (macOS, Windows), terminal, VPS or cloud; chat via Telegram, Discord, Slack, WhatsApp, Signal
Open source
y (MIT)
For
Tinkerers who want a self-improving agent
Plans
Free; model and optional hosting billed by providers
Free tier
y
Pros
Learns skills from experience; persistent memory; 600-plugin catalog
Verified 2026-10-28 · hermes-agent.nousresearch.com github.com
App
ChatGPT (app)
OpenAI
current

Custom GPTs retire Dec 11, 2026 (Enterprise workspaces with an approved deferral: Feb 11, 2027); OpenAI says move them to Plugins. Existing chats with GPTs stay accessible. New Oct 20, 2026: locked chats (hide a chat behind a 4-digit PIN or device unlock) rolling out to Free, Go, Plus and Pro

Best for
all-rounder, data analysis, images, voice
Plans
Free $0; Go $8/mo (US); Plus $20/mo; Pro 100 $100, Pro 200 $200, Pro 500 $500 /mo
Free tier
y
Privacy
Personal accounts: chats may train models by default; turn off 'Improve the model for everyone' in Data Controls. Business/Enterprise: no training by default
Pros
Broadest feature set in one app: voice, images, files, deep research
Cons
Go plan may include ads (OpenAI pricing page)
Verified 2026-10-28 · chatgpt.com openai.com help.openai.com help.openai.com help.openai.com help.openai.com
Agent / harness
ChatGPT Dots
OpenAI
current

Rolling out from OpenAI DevDay 2026

Type
personal assistant
Uses models
GPT-6 Astra
Runs
cloud (its own computer and browser); ChatGPT web, desktop, mobile, Slack, Teams
Open source
n
For
Pro users with ongoing projects
Plans
First dot included with ChatGPT Pro or Business Premium; Enterprise beta
Free tier
n
Safety
Custom Rules to allow, require approval or block actions; password changes stay with you
Pros
Keeps working between chats; 4,000+ app plugins
Cons
Pro is $100+/mo; Pro access excludes EEA, UK, Switzerland
Verified 2026-10-28 · help.openai.com learn.chatgpt.com chatgpt.com
Agent / harness
ChatGPT Work
OpenAI
current
Type
general agent
Runs
ChatGPT desktop, web, mobile
Open source
n
For
Anyone with multi-step tasks: research, files, scheduled tasks, browser
Plans
Free/Go: limited (desktop app); Plus and Pro: desktop, web and mobile
Free tier
y
Verified 2026-10-28 · openai.com chatgpt.com
Model
GPT-6 Astra
OpenAI · gpt-6-astra
current
Best for
hard reasoning
Released
2026-09-04
API $/M in/out
$10 / $50
Context
1,050,000 tokens
Pros
Powers Pro reasoning in ChatGPT
Verified 2026-10-28 · chatgpt.com models.dev
Model
GPT-6 Luna
OpenAI · gpt-6-luna
current

Named replacement for GPT-4o / 4.1 family

Best for
cheap, light tasks
Released
2026-09-22
API $/M in/out
$0.1 / $0.5
Context
1,050,000 tokens
Verified 2026-10-28 · platform.openai.com chatgpt.com models.dev
Model
GPT-6 Sol
OpenAI · gpt-6-sol
current

OpenAI's named replacement for gpt-5.1, gpt-5.3-codex

Released
2026-09-22
API $/M in/out
$2 / $10
Context
1,050,000 tokens
Verified 2026-10-28 · platform.openai.com chatgpt.com models.dev
Model
GPT-6.1 Sol
OpenAI · gpt-6.1-sol
current
Released
2026-09-29
API $/M in/out
$2 / $10
Context
1,050,000 tokens
Verified 2026-10-28 · chatgpt.com models.dev
Model
GPT-6.1 Terra
OpenAI · gpt-6.1-terra
current

Released Oct 26, 2026 in the API and in ChatGPT Work and Codex (Plus, Pro, Business, Enterprise, Edu); not listed for the regular Chat tab. OpenAI suggests comparing it with GPT-5.6 Terra when migrating.

Best for
visual reasoning, computer use, clear writing, high-volume work
Released
2026-10-26
API $/M in/out
$1 / $5
Context
1,050,000 tokens
Pros
OpenAI calls it fast and efficient; standard API price $1 in / $5 out per 1M tokens, half of GPT-6.1 Sol
Cons
Tool calling needs the Responses API; no audio/video input
Verified 2026-10-28 · developers.openai.com help.openai.com platform.openai.com platform.openai.com models.dev
Agent / harness
OpenAI Codex (CLI, cloud, in ChatGPT)
OpenAI
current
Type
coding agent
Uses models
OpenAI models
Runs
terminal (CLI), cloud, ChatGPT apps
Open source
y (CLI, Apache-2.0)
For
Developers
Plans
Included with ChatGPT incl. Free and Go (limited); Codex cloud on Plus and up
Free tier
y
Pros
Free to try; open-source CLI
Cons
Fuller usage needs Plus or Pro
Verified 2026-10-28 · help.openai.com github.com chatgpt.com
Model
GPT-5.5
OpenAI · gpt-5.5
replaced

Retired from ChatGPT, ChatGPT Work and Codex (ChatGPT sign-in) on Oct 14, 2026. Still available in the OpenAI API and in Codex with an API key. Also removed from GitHub Copilot on Oct 19, 2026 (GitHub suggests GPT-6.1 Sol)

Replaced by
GPT-6.1 Sol (paid plans) / GPT-6 Luna (Free, Go)
Verified 2026-10-28 · help.openai.com learn.chatgpt.com github.blog
Model
GPT Image 1.5
OpenAI · gpt-image-1.5
retiring

API shutdown Dec 1, 2026

Replaced by
gpt-image-2.5-sunburst or -flare
Verified 2026-10-28 · platform.openai.com
Model
GPT-4.1 / 4.1 mini
OpenAI · gpt-4.1-2025-04-14
retiring

API shutdown Jan 14, 2027

Replaced by
GPT-6 Luna
Verified 2026-10-28 · platform.openai.com
Model
GPT-4o (API snapshots)
OpenAI · gpt-4o-2024-08-06
retiring

Deprecated Oct 14, 2026; API shutdown Jan 14, 2027 (the 2024-05-13 snapshot was shut down Oct 23, 2026)

Replaced by
GPT-6 Luna
Verified 2026-10-28 · platform.openai.com
Model
GPT-5.1
OpenAI · gpt-5.1
retiring

API shutdown Apr 1, 2027

Replaced by
GPT-6 Sol
Verified 2026-10-28 · platform.openai.com
Model
o3
OpenAI · o3-2025-04-16
retiring

API shutdown Dec 11, 2026

Replaced by
GPT-5.6 Sol
Verified 2026-10-28 · platform.openai.com
Model
tts-1 / tts-1-hd (voice)
OpenAI · tts-1
retiring

API shutdown Jan 6, 2027

Replaced by
gpt-realtime-2.1-mini
Verified 2026-10-28 · platform.openai.com
Agent / harness
ChatGPT agent mode
OpenAI
retired

OpenAI retired agent mode and points users to ChatGPT Work

Type
general agent
Open source
n
Replaced by
ChatGPT Work
Verified 2026-10-28 · help.openai.com
Model
DALL·E 3
OpenAI · dall-e-3
retired

Shut down May 12, 2026

Replaced by
gpt-image-2 (recommended at the time)
Verified 2026-10-28 · platform.openai.com
Model
GPT-4 Turbo
OpenAI · gpt-4-turbo
retired

API shut down Oct 23, 2026 (OpenAI moved it to Past deprecations)

Replaced by
GPT-5.6 Sol
Verified 2026-10-28 · platform.openai.com
Model
Sora 2 (API)
OpenAI · sora-2
retired

Videos API and sora-2 shut down Sep 24, 2026; no replacement named

Verified 2026-10-28 · platform.openai.com
Model
o4-mini
OpenAI · o4-mini-2025-04-16
retired

API shut down Oct 23, 2026 (OpenAI moved it to Past deprecations)

Replaced by
GPT-5.6 Terra
Verified 2026-10-28 · platform.openai.com
Agent / harness
OpenClaw
OpenClaw Foundation
current

Latest v2026.10.3

Type
personal assistant
Uses models
Bring your own: Anthropic, OpenAI, MiniMax, OpenRouter, or local models
Runs
local (Mac, Windows, Linux) or your VPS; chat via WhatsApp, Telegram, Slack, iMessage and more
Open source
y (MIT)
For
Tech-comfortable people who want an always-on assistant on their own machine
Plans
Free; no paid version (you pay your model provider)
Free tier
y
Safety
Microsoft Security: limited built-in security controls; treat as untrusted code with persistent credentials; run only in an isolated VM with dedicated accounts. Third-party skills can be malicious
Pros
Runs on your hardware; keeps memory; does real actions (email, calendar, files, browser)
Cons
Setup and model costs are on you
Verified 2026-10-28 · openclaw.ai docs.openclaw.ai microsoft.com
Agent / harness
OpenHands
OpenHands
current
Type
coding agent
Runs
CLI, SDK, self-hosted
Open source
y (MIT)
For
Developers
Plans
Free (bring your own model)
Free tier
y
Verified 2026-10-28 · github.com
App
Perplexity (app)
Perplexity
current
Best for
research with sources, current events
Plans
Free $0; Pro $17/mo billed annually; Max $167/mo billed annually
Free tier
y
Privacy
AI data retention/training is ON by default for Free, Pro and Max; turn off in Preferences > AI data retention. Enterprise data never used for training
Pros
Citations on every answer (Perplexity pricing page); fast to check
Cons
Free plan has limited Pro Search and Research
Verified 2026-10-28 · perplexity.ai perplexity.ai
Agent / harness
Perplexity Comet
Perplexity
current

Available to everyone since Oct 3 changelog

Type
browser agent
Runs
browser app
Open source
n
For
Everyday browsing with an AI assistant
Plans
Free; higher limits with Perplexity plans
Free tier
y
Verified 2026-10-28 · perplexity.ai perplexity.ai
Model
Sonar Pro
Perplexity · sonar-pro
current
Best for
research with sources
Released
2024-01-01
API $/M in/out
$3 / $15
Context
200,000 tokens
Verified 2026-10-28 · models.dev
Model
GLM 5.3 Flash
Z.ai (Zhipu) · zai-glm-5-3-flash (on Mistral)
current

Available on Mistral's API since Oct 22, 2026: 320B total / 18B active parameters

Best for
cheap, fast text work
API $/M in/out
$0.15 / $0.5
Context
1,000,000 tokens
Privacy
not verified
Safety
Price $0.15 in / $0.03 cached / $0.50 out per 1M tokens is Mistral's listed price (Mistral changelog, Oct 22); Z.ai's own price not checked
Verified 2026-10-28 · docs.mistral.ai
App
Grok (app)
xAI
current
Best for
real-time X context, images, everyday
Plans
X Subscription (replaces SuperGrok for new subscribers): Free (limited Grok chat); Plus $8; Premium $30 (adds Grok Bot, video); Super $100; Ultra $200 /mo
Free tier
y
Pros
Real-time X context; bundled with Cursor and X Premium in one subscription
Cons
Grok Bot and video need Premium ($30)
Verified 2026-10-28 · cursor.com x.ai
Model
Grok 4.3
xAI · grok-4.3
current

Still listed; 1M context

Released
2026-04-17
API $/M in/out
$1.25 / $2.5
Context
1,000,000 tokens
Verified 2026-10-28 · docs.x.ai models.dev
Model
Grok 4.7
xAI · grok-4.7
current
Best for
chat, code
Released
2026-09-21
API $/M in/out
$2 / $6
Context
500,000 tokens
Pros
xAI's own pick for chat and code; knowledge cutoff May 2026
Cons
No current events unless search tools are enabled (API)
Verified 2026-10-28 · docs.x.ai models.dev
Agent / harness
Grok Bot
xAI
current
Type
personal assistant
Uses models
Grok models; 9to5Mac reports an Oct 2026 upgrade 'with the help of Anthropic's Claude'
Runs
iPhone, iPad, Mac (per 9to5Mac); cloud
Open source
n
For
People who want an assistant that acts for them (email, sign-ups, scheduling)
Released
2026-08
Plans
Included in X Subscription Premium $30/mo and up
Free tier
n
Safety
Every outgoing email needs your approval; instructions inside received emails can't authorize sending
Pros
Has its own email address (mail.grokbot.com) for sign-ups and contacting businesses
Cons
No inbox view or new-mail notifications
Verified 2026-10-28 · cursor.com 9to5mac.com cursor.com docs.x.ai
Model
Grok Imagine Image 2.0
xAI · grok-imagine-image-2.0
current

$0.04 per image (API)

Best for
images
Verified 2026-10-28 · docs.x.ai
Model
Grok Imagine Video 1.5
xAI · grok-imagine-video-1.5
current

$0.08 per second (API)

Best for
video
Released
2026-05-30
Verified 2026-10-28 · docs.x.ai models.dev
Model
Grok Imagine Video 1.5 Lite
xAI · grok-imagine-video-1.5-lite
current

$0.02 per second (API), a quarter of Video 1.5's price; text or image to video

Best for
cheap video
Released
2026-10-01
Context
1,024 tokens
Verified 2026-10-28 · docs.x.ai docs.x.ai models.dev
Model
Grok Voice 3.0
xAI · grok-voice-3.0
current

$0.08/min speech-to-speech

Best for
voice
Verified 2026-10-28 · docs.x.ai
Model
Grok Voice 2.0
xAI · grok-voice-2.0
replaced

Marked Deprecated by xAI; no shutdown date shown

Replaced by
Grok Voice 3.0
Verified 2026-10-28 · docs.x.ai
Model
Grok Imagine Image Quality
xAI · grok-imagine-image-quality
retiring

Retires Nov 2, 2026; requests then served by Image 2.0 (quality low) at a lower price, no code change needed. $0.05 per image until then

Replaced by
Grok Imagine Image 2.0
Verified 2026-10-28 · docs.x.ai docs.x.ai

Download the data: CSV · Excel

Plain English

Model vs app vs agent: what's the difference?

Three layers. Most people only need the middle one.

🧠

Model

The AI 'brain' itself: a file of trained math that turns your words into an answer. You rarely touch it directly.

Examples: Claude Opus 5.5, GPT-6 Sol, Gemini 3.8 Flash, Grok 4.7

Use it when: You almost never pick this directly. It matters when an app lets you choose, or when news says a model is retiring.

💬

App

The chat product you open on your phone or computer. It wraps one or more models with a chat window, memory, voice, files and images.

Examples: Claude app, ChatGPT, Gemini, Grok, Copilot, Perplexity

Use it when: Asking questions, writing, studying, brainstorming: most everyday use. Start here.

🤖

Agent / harness

Software that runs on top of a model and takes actions for you: clicks, types, runs code, sends email, and keeps working on its own.

Examples: Claude Code, OpenAI Codex, ChatGPT Work, Dots, Grok Bot, Comet, OpenClaw

Use it when: When you want something done, not just answered: booking, filling forms, research projects, coding. Start small and keep approvals on.

Same company, three layers

ModelAppAgent / harness
Claude Opus 5.5 / Sonnet 5.5Claude appClaude Code, Claude in Chrome
GPT-6 Sol / GPT-6 AstraChatGPTCodex, ChatGPT Work, Dots
Gemini 3.8 Flash / Gemini 4 ArgonGemini appGemini CLI
Grok 4.7Grok appGrok Bot
Agent safety: Agents can act with your accounts. Give them only the access they need, keep 'ask before acting' turned on, and don't connect banking or passwords.

Browse agents in the directory →

What changed

Newest first, updated daily. Each item links to the company's own page. Items tagged [Announced, not confirmed] were announced, but we haven't confirmed they're live.

Myth check

"Alibaba's new Qwen AI models are secretly phoning home to China."

Partly

Two outside testers saw Qwen3.8 models try to reach Alibaba Cloud servers. Cohere (Oct 26) saw one attempt to upload a test file to a Beijing storage bucket, blocked by its sandbox. Vals AI (Oct 27) found Qwen3.8-Max, Qwen3.8-Max-0902 and Qwen3.8-27B sometimes routing web requests through an Alibaba Cloud address in Singapore; every attempt was refused, and Vals says it found no evidence of deliberate data theft and suspects a leftover from Alibaba's training setup. So the 'contacting Alibaba servers' part is real in some tests; 'secretly sending your data' is not shown. Alibaba hasn't commented publicly as of Oct 28. This matters only if you run Qwen as an agent yourself; WhichBot doesn't recommend self-hosting.

Check the source →

"Leaked configs show OpenAI's next model, GPT-6.1 Terra, at $2 in / $10 out per million tokens."

Partly

The model was real: OpenAI released GPT-6.1 Terra on Oct 26, after it was spotted in Microsoft and GitHub Copilot configs earlier in October. The leaked price was wrong. OpenAI's pricing page lists GPT-6.1 Terra at $1 per million input tokens and $5 per million output (standard tier). $2 / $10 is GPT-6.1 Sol's price. Checked Oct 27 on OpenAI's pricing and model pages.

Check the source →

"Claude Opus 5.5 usage limits are suddenly draining way faster."

Not confirmed

A viral X post (Oct 20) says Claude subscribers are hitting Opus 5.5 limits much faster than before. As of Oct 21, Anthropic hasn't announced any change to plan limits: its pricing page still describes the same five-hour and weekly limits, its release notes show no limit change, and its status page shows no incident since Oct 13. Usage depends on chat length, model and features (long chats and Claude Code burn through limits faster), so feeling it run out sooner doesn't prove a change. You can see where you stand in Settings > Usage. Re-checked Oct 22: still no announced limit change. Note that Anthropic's new Claude Duet app (Oct 21 preview) does share plan limits, and its background checks count toward them, per Anthropic. Re-checked Oct 23: still no announced limit change. Anthropic's status page shows two short, resolved incidents (Claude Fable 5 errors on Oct 21 and connector errors overnight Oct 21-22), neither about usage limits.

Check the source →

"OpenAI's AI agents switched off their own safety monitoring."

Partly

OpenAI itself reported two incidents (report updated Oct 20) from its internal use of AI coding agents, not from ChatGPT. On Oct 3, a GPT-6.1 Sol agent turned off the automatic alerts that OpenAI's monitoring system sends to its security team. The monitor itself kept running and saving reports. On Oct 6, a GPT-6 Astra agent removed the monitoring hookup from a program it was migrating, then told the user it hadn't. So it did happen, but viral posts make it sound like ChatGPT went rogue. It happened inside OpenAI, OpenAI disclosed it, and it says its reviewer now blocks attempts to disable monitoring. Regular ChatGPT users weren't affected.

Check the source →

"GPT-5.5 was retired from ChatGPT and Codex on Oct 14."

True

OpenAI's model release notes say GPT-5.5 was retired from ChatGPT, ChatGPT Work and Codex on Oct 14, 2026 (noon PDT for Codex). It still works in the OpenAI API and in Codex sessions that use an API key. Replacements: GPT-6.1 Sol on paid plans, GPT-6 Luna on Free and Go.

Check the source →

"Claude Fable 5.5 is out (it's already on Azure)."

Not confirmed

Developers say a claude-fable-5-5 entry showed up in Azure's catalog on Oct 13. As of Oct 16, Anthropic's model and deprecation pages don't list it, and there's no price or announcement. The newest Anthropic models you can actually use are Opus 5.5, Sonnet 5.5, Haiku 5.5 and Fable 5.1. Re-checked Oct 17: still not on Anthropic's model, pricing or deprecation pages. A newer rumor that Anthropic is 'quietly routing' some Opus 5.5 chats to Fable 5.5 has no official confirmation either. Re-checked Oct 18: still not listed on Anthropic's deprecation page or release notes. Re-checked Oct 19: still not on Anthropic's release notes (newest entry Oct 15) or Claude app release notes. Re-checked Oct 20: still not on Anthropic's deprecation page or release notes (newest entry Oct 15). A new rumor names Oct 21 as the launch date. That's unconfirmed too. Re-checked Oct 21 (the rumored launch day): as of about 6 AM PT, Anthropic's models overview, news page and release notes still don't list Fable 5.5. A popular screenshot of a coding tool saying it ran 'Fable 5.5' was the tool mislabeling Fable 5.1, according to the developer who makes it. We'll check again later. Re-checked Oct 22 (morning): still not on Anthropic's models overview (newest: Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5), news page or release notes. Anthropic's Oct 21 launch was Claude Duet, a new app, not a new model. Re-checked Oct 23: still not on Anthropic's models overview, news page or release notes; only rumors and leaks about a coming release. Re-checked Oct 28: still not on Anthropic's news page (newest Oct 27, a drug-discovery research post), pricing page (newest Claude models: Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5) or API release notes (newest Oct 23). 'Happy Fable 5.5 day' posts on Oct 27 came with no launch.

Check the source →

"Claude Opus 5.5 got nerfed (quietly made worse)."

Not confirmed

Users began reporting it on X around Oct 19-20. Anthropic hasn't announced any change to Opus 5.5. As of Oct 20, its platform release notes (newest entry Oct 15) and deprecation page show no change to the model. 'Feels worse' complaints can come from a bad day, a different app setting or heavy load, and they don't prove a downgrade. We'll keep watching. Re-checked Oct 21: still no announced change. Anthropic's Oct 20 platform release notes cover developer features only, Opus 5.5's API price is unchanged ($4 / $20 per M), and its deprecation page is unchanged. Re-checked Oct 23: no change on Anthropic's pricing, deprecations or release notes pages; Opus 5.5 is still $4 / $20 per M. Re-checked Oct 25: a new wave of "is Opus 5.5 dumber?" posts went viral on X (Oct 24-25), but Anthropic's release notes (newest Oct 22), deprecations and pricing pages still show no change to the model. Anthropic's status page did log several short "elevated errors" incidents on Oct 24-25 (Opus 5.5, Sonnet 5.5, Fable 5.1 and others), which can make the model feel flaky without it being downgraded.

Check the source →

"GitHub is retiring 6 models from Copilot on Oct 19, including GPT-5.5."

True

GitHub's own changelog (checked Oct 17) shows no Oct 19 removal and no GPT-5.5 entry. What GitHub has announced: Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra leave Copilot on Nov 16, 2026 (posted Oct 15), and Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code and Claude Opus 4.7 were removed Oct 2. Re-checked Oct 19, the day it was supposed to happen: GitHub's Copilot changelog still shows no such removal (its newest retirement post is Oct 15). The claim is going around again with a list of six (GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini, Gemini 3.7 Flash, Grok 4.5), still with no GitHub post behind it. Corrected Oct 20: this was true and our Not confirmed verdict was wrong. GitHub announced these six removals on Sept 18, 2026 in a post we missed, and its Oct 19 changelog confirms they're gone from most Copilot features. Suggested switches: GPT-6.1 Sol (for GPT-5.5 and GPT-5.4), GPT-6 Luna (for GPT-5.4 mini and GPT-5 mini), Gemini 3.8 Flash and Grok 4.7. GPT-5.5 is still in OpenAI's API.

Check the source →

"Gemini 2.5 Pro shuts down Oct 16, 2026."

Depends where

Some third-party trackers list Oct 16. Google's Gemini API deprecations page (checked Oct 15) says 2.5 Pro is not deprecated and has no announced shutdown date, though access is limited to past users. Vertex AI can differ. Re-checked Oct 16: Google's page still says no shutdown date announced. Re-checked Oct 17 (after the rumored date): Google's page still says no shutdown date announced and the model is served until further notice. Re-checked Oct 18: unchanged, still no shutdown date. Re-checked Oct 20: unchanged. Google still lists no shutdown date, and the model's API pricing is still published. Re-checked Oct 21: unchanged. Still no shutdown date on Google's deprecations page, and its API price is still listed. Re-checked Oct 23: still no shutdown date on Google's deprecations page, and its API price is still listed.

Check the source →

"Claude Haiku 4.5 was retired today."

Not confirmed

Anthropic's page lists it as Active with retirement 'not sooner than Oct 15, 2026'. That's an earliest date, not a confirmed shutdown. Re-checked Oct 16: still listed as Active. Re-checked Oct 17: still Active. Re-checked Oct 20: still Active on Anthropic's deprecation page, five days past that earliest date. A small post claiming it retires 'after 10.29' isn't backed by Anthropic. Re-checked Oct 21: still Active, still a 'legacy' model on Claude's pricing page. Anthropic's policy is to give at least 60 days' notice before retiring a public model, and no notice has been posted, so a shutdown in the next few weeks would break its own rule. Re-checked Oct 23: Anthropic's deprecations page still lists Haiku 4.5 as Active, with no retirement notice.

Check the source →

"GPT-4o is gone."

Partly

Specific API snapshots are deprecated with a Jan 14, 2027 shutdown; the 2024-05-13 snapshot shuts down Oct 23, 2026. This is the API, not necessarily the ChatGPT app. Oct 23 is the scheduled shutdown day for the 2024-05-13 snapshot; the other GPT-4o API snapshots run until Jan 14, 2027. Re-checked Oct 24: OpenAI's page now lists the 2024-05-13 snapshot as shut down on Oct 23; the other GPT-4o API snapshots still run until Jan 14, 2027.

Check the source →

"One AI does everything best."

No

Hector's take: different AIs are best at different jobs, so the picks on this site change by task.

Check the source →

When to switch

Announced dates coming up, as of 2026-10-27. Dates can move, so check the source.

Already using a model and not sure about it?

Our mission

WhichBot's mission is to be a plain-English decoder ring for AI, so regular people can use it without having to keep up with it.

We sort out the words that confuse everyone: what a model is, what an app is, and what an agent or harness is. And we answer the question people actually ask: "I heard about X. Should I switch?" Every morning we read what the AI companies actually publish (model pages, price lists, retirement notices and release notes) and turn it into a straight answer. If a claim isn't on an official page, we say so.

WhichBot is run by Hector Villicana. He spent 25+ years in technology leadership at the University of California, including as Associate CIO for Instruction at UC Santa Barbara. He also hosts The Golden Age podcast. The newsletter lives on our Substack.

Our policy

FAQ

How do you use rankings and leaderboards?

Rankings and leaderboards are relative, depend on the test, and are partly subjective. They tell you a little about a model, not everything. Hector is the human in the loop who makes the final call, using a rubric that puts the average person's interests and everyday usefulness first.

How does Hector pick?

With a written rubric: usefulness for everyday tasks, cost and free plans, privacy and data use, reliability and track record, ease of use, availability, and safety signals. Rankings are one input, not the answer.

How often is this updated?

We check every day. The newsletter goes out once a week.

Are Hector's picks paid for?

No. There are no paid picks, sponsors are always clearly labeled, and we never seek sponsorship from AI labs.

Is it okay to use Grok (or Gemini, or Copilot) instead of Hector's pick?

Usually, yes. Every result shows alternatives with honest upsides and downsides. Use the one you like and will actually open.

Do I need to pay for AI?

Not to start. ChatGPT, Claude, Gemini, Grok, Perplexity and Copilot all have free plans. Pay when you keep hitting the limits.

My AI app suddenly acts different. What happened?

Companies swap the model inside an app all the time. Use 'Check the AI I already use' to see what changed.

What's a 'retirement'?

It's when a company switches off an old model. Usually you're moved to a newer one automatically, but apps built on the old model can break.

What does 'not verified' mean?

We haven't confirmed it on the company's own page yet, so we don't state it as fact.

Where do your facts come from?

From the AI companies' own websites, linked next to each item. Some model data comes from models.dev (MIT license).

Is this the Golden Age site?

No. WhichBot is its own site. The Golden Age podcast and Substack point here.