Model Updates

New and updated AI models from every major provider.

This Week

Perplexitymodel

Link·Aug 13, 2026

Grok 4.6 in Perplexity and Perplexity Computer

xAI's Grok 4.6 is now selectable in Perplexity and in Perplexity Computer.

Best for: running the same Perplexity work on a cheaper model without giving up much quality
When to use it: cost per task matters more to you than squeezing out the last bit of quality
Who benefits: Perplexity and Perplexity Computer users who pick their own model

Googlemodel

Link·Aug 13, 2026

Gemini 3.7 Flash

Google's new workhorse Gemini model for coding and agent work, priced at half of 3.6 Flash through the end of 2026.

Best for: multi-step agent and coding work you were sending to a pricier model for reliability
When to use it: a task needs planning and tool calls across several steps, not one answer
Who benefits: developers on the Gemini API, and Gemini app users through Spark

Perplexitymodel

Link·Aug 12, 2026

Nemotron 3.5 Lightning in the Perplexity Agent API

NVIDIA's open-weight Nemotron 3.5 Lightning is now available in Perplexity's Agent API, priced for the high-volume, repetitive parts of an agent's work.

Best for: The cheap, repetitive steps inside an agent: calling tools, checking results, and handing work off to subagents.
When to use it: An agent is making the same kind of call thousands of times and a frontier model's price stops making sense.
Who benefits: Developers running always-on agents where the constraint is cost per call, not reasoning depth.

Microsoft Copilotmodel

Link·Aug 11, 2026

Anthropic Models in Copilot for Word

Copilot in Word lets you pick an Anthropic model instead of an OpenAI one when editing a document.

Best for: Long-form document editing where you want a second model's read on the same draft.
When to use it: The edits an OpenAI model returns are not landing and you want to run the same passage through a different one.
Who benefits: Content and communications people drafting substantial documents in Word rather than short replies.

OpenAImodel

Link·Aug 10, 2026

GPT-5.6-Cyber

OpenAI's cybersecurity-specific model, available only to defenders vetted into the new Daybreak Red access tier.

Best for: Authorized vulnerability research, exploit validation, and security testing inside an approved security team.
When to use it: You are a vetted defender who needs a model that engages with dual-use security work instead of refusing it.
Who benefits: Security researchers and red teams at organizations approved into OpenAI's Daybreak program.

Last Week

Anthropicmodel

Link·Aug 7, 2026

Claude Fable 5 Biology Safeguards Update

Fable 5 now answers everyday health and biology questions it used to refuse or hand off.

Best for: Health, medical, and science questions that Fable 5 was declining to answer
When to use it: You hit a refusal on an ordinary health or biology question and had to reword it or switch models
Who benefits: Anyone using Fable 5 for health, science, or educational work

Perplexitymodel

Link·Aug 6, 2026

GPT-5.6 Terra and Luna in Perplexity Computer

Perplexity Computer swaps in Terra as the default model for subagents and Luna as the model behind scheduled automations.

Best for: Running multi-step agent work in Perplexity Computer without picking a model for each subagent.
When to use it: You already run Computer subagents or scheduled automations and want the newer models applied by default.
Who benefits: Perplexity Computer users running recurring research or operations workflows.

OpenAImodel

Link·Aug 6, 2026

GPT-5.6 Sol and Luna in ChatGPT

ChatGPT's chat experience gets an updated GPT-5.6 Sol for paid users and unlimited GPT-5.6 Luna chats for free users.

Best for: Everyday ChatGPT work where you want a direct answer instead of a padded one, and where you sometimes need it to think harder.
When to use it: You are on Plus or Pro and want to dial reasoning up or down per question, or you use the free tier and keep hitting message caps.
Who benefits: Marketers who live in ChatGPT chat all day, and teams whose members are still on the free plan.

July 2026

OpenAImodel

Link·Jul 31, 2026

GPT-5.4 and GPT-5.4 mini Retired from ChatGPT

Both models stop working for anyone signed in with a ChatGPT account on August 31, 2026. They stay available through the API.

Best for: Auditing where you still call GPT-5.4 before the date passes.
When to use it: Saved prompts, custom GPTs, or Codex sessions are pinned to GPT-5.4 and need repointing.
Who benefits: ChatGPT and Codex users with anything hardcoded to the 5.4 family.

Perplexitymodel

Link·Jul 28, 2026

Kimi K3 in Perplexity

Kimi K3 is now one of the models you can pick inside Perplexity and Perplexity Computer, on Pro and Max.

Best for: Putting a different model on a task without leaving Perplexity and rebuilding your prompt somewhere else.
When to use it: Your usual model keeps stalling on a specific kind of task and you want a second engine on it.
Who benefits: Pro and Max subscribers who already switch models depending on the job.