This Week in AI Tools: Cheaper Haiku, GPT-6 Rollout

Two announcements landed on 7 October, and between them they change what a free account can do and what a cheap API call costs. Neither is as simple as its headline. Here is this week in AI tools, 4 to 10 October, with the small print that the headlines left out.

OpenAI puts GPT-6 in front of free users

OpenAI announced on 7 October that GPT-6 is rolling out across ChatGPT, split by tier: GPT-6 Sol to Plus, Pro, Business and Enterprise, and GPT-6 Luna to Free and Go from the following day. It also introduced Intelligent UI, which lets a reply include graphics, buttons, forms and charts rather than only prose.

OpenAI has published no free-tier message limit, which is the detail that decides whether this is useful to you. We went through the tiers, the API prices and the 272K context cliff separately in is GPT-6 free.

Anthropic cuts Haiku to $0.10 per million tokens — up to 100,000 of them

Also on 7 October, Anthropic introduced Claude Haiku 5.5, calling it its fastest and cheapest model to date. The published rates:

	Prompts under 100k tokens	Prompts over 100k
Input	$0.10 / M tokens	$0.50 / M tokens
Output	$0.50 / M tokens	$2.50 / M tokens
Cache reads	$0.01 / M tokens	$0.05 / M tokens

That 100,000-token line is the detail to carry away: every rate multiplies by five above it. Anthropic reports 72.4% on OSWorld 2.1 for computer use, 39.2% on Terminal-Bench 4.0 for agentic coding, and early testers seeing latency fall by more than 30%.

Two things complicate the headline. Anthropic's own migration guide states the new tokenizer produces roughly 30% more tokens for the same text than Haiku 4.5 did, which raises real cost and brings that 100k cliff closer than a character count suggests. And on Artificial Analysis, the maximum-reasoning configuration generated 440M tokens across the Intelligence Index against a median of 100M, while the medium setting came in at 54M. The cheapest model per token is not automatically the cheapest per task, and the effort level does most of the work.

Anthropic also halved Sonnet 5.5 cache reads to $0.10 per million — about 20% off agentic task costs, it says — and added monthly API credits: $100 for Max 5x, $200 for Max 20x, up to $500 pooled for Team. Our Claude review and ChatGPT vs Claude compare the two on everyday work rather than benchmarks.

A free year of Claude Team for young companies

On 6 October, during SF Tech Week, Anthropic expanded Claude for Startups: a free year of Claude Team with up to five premium seats, $1,000 in API credits, access to Claude Marketplace for building plug-ins, and virtual office hours with its Applied AI team. TechCrunch reported the eligibility test as founded within the last five years or funded within the last two — one or the other, not both.

ElevenLabs ships agent branching

Quieter and firmly developer-facing: on 5 October ElevenLabs put merge proposals into alpha, letting a team review configuration changes and test results before merging an agent branch, and made structured procedures compile on save. The SDKs added gpt-6.1-sol and claude-sonnet-5-5 as agent voice models.

If you use ElevenLabs to make audio rather than to build agents, none of this affects you — our ElevenLabs review covers what does.

The rest of this week in AI tools

Anthropic published a 2026 Usage Policy update on 8 October, effective 12 November 2026. Two changes matter to developers: operators of AI-connected hardware that could cause physical harm must supervise it and be able to stop it, and region restrictions are stated to apply to where a user is physically located and to majority-owned entities, not only to where a company is registered.

What we would actually do about it

If you use ChatGPT free, you have a better model this week at no cost and no action required. If you pay for Plus mainly for model quality, wait for published Luna rate limits before cancelling — the gap between tiers may now be quota rather than capability.

If you build on Haiku, measure rather than assume. Price the work in tokens at your real prompt lengths and your chosen effort level, then compare against Haiku 4.5. A 90% rate cut and a 30% token increase do not net out to 90%, and Anthropic puts the practical average nearer 75%.

FAQ

Is GPT-6 free?

A version of it is. OpenAI said on 7 October that GPT-6 Luna began rolling out to ChatGPT's Free and Go tiers, while GPT-6 Sol goes to Plus, Pro, Business and Enterprise. OpenAI has not published usage limits for Luna on the free tier, so how many messages you get is not yet known.

How much does Claude Haiku 5.5 cost?

$0.10 per million input tokens and $0.50 per million output tokens for prompts under 100,000 tokens. Above that threshold every rate is five times higher: $0.50 and $2.50. Cache reads are $0.01 per million under the threshold.

Is Haiku 5.5 actually 90% cheaper than Haiku 4.5?

Not in practice. Anthropic's migration guide notes the new tokenizer counts about 30% more tokens for the same text, and Anthropic itself puts the average saving nearer 75%. Heavy reasoning settings push real cost higher again.

What is ChatGPT's Intelligent UI?

A change to how answers are presented rather than to the model. Responses can include graphics, buttons, forms, charts and small interactive tools built in the conversation, with the format chosen to suit the question.

Who qualifies for the free year of Claude Team?

Per TechCrunch's report of the 6 October announcement, a company founded within the last five years or funded within the last two. It covers one year of Claude Team with up to five premium seats plus $1,000 in API credits.