Skip to main content
Radar

Five or fewer confirmed AI changes, with original sources, on mornings when something changed.A free source-linked morning brief.

Start free brief
BenchLM

Sent every week

The model changes worth your attention

Rank moves, price changes, launches, and one clear decision note. One email every week.

Get the next weekly brief

Read the sample below first. Subscribe when the format earns the space in your inbox.

Join 2,000+ readers.

What arrives

Useful change, not a news firehose

Each issue is built for people choosing, buying, or operating language models.

Rank moves with the evidence attached

See who moved, what changed, and whether the public benchmark coverage supports the move.

Material API price changes

Track pricing that could change a production shortlist, without filling your inbox with every minor update.

Release and availability context

Separate a launch headline from the access limits, evidence gaps, and reasons to wait.

One decision note

Leave with a model to consider, a lower-cost alternative, or a clear reason not to switch yet.

Real sample issue

See the whole format

This is the editorial content of a sent issue, adapted from email to the web.

Weekly LLM benchmark digest

OpenAI is keeping ZDR. The safety system is moving.

On August 19, OpenAI restated Zero Data Retention for eligible API customers: prompts and responses are not retained after a request, staff do not review that content, and enterprise data is not used for training unless the customer opts in. Frontier models now run longer, multi-turn agentic work, where some risks only emerge across a sequence. OpenAI is previewing Private Safety Processing so ZDR can survive that requirement. For ZDR deployments, content remains on infrastructure the customer controls. OpenAI is also developing customer-controlled-key storage on its infrastructure, with automated systems returning narrow safety signals rather than the underlying prompts or responses.

What it means

ZDR remains a procurement optionAsk whether you are eligible, where content is stored, who holds the keys, and which signals leave your boundary.Multi-turn safety is now a productIf an agent keeps acting after a stop, that is the class of risk Private Safety Processing is designed to catch.The CSAM exception remainsImages flagged as possible CSAM are still retained for manual review and reporting, including in ZDR deployments.A preview is not a signed controlPrivate Safety Processing is in early-customer tests. OpenAI plans the technical white paper and rollout for September, while customer-controlled-key storage remains in development.

The limit

We have not run this system. The public note does not specify eligibility, regional availability, or the exact signal schema. If you already use ZDR, wait for the September details before rewriting the DPA. This kind of policy change also will not appear as a new model ID.

Analysis worth opening

Fast on a wafer is not a shortlistCerebras is quicker than Groq on the models both host. The decision is which weights sit on the wafer, and whether an agent is actually waiting on tokens.Groq is not a modelAfter Nvidia licensed the chip and hired the team, what remains is a fast, narrow catalog. Most people searching “groq” still think they are buying a brain.K3 is open. You still cannot run it.Moonshot’s 2.8T flagship sits next to the closed frontier. Open means downloadable shards and a license, not a machine you own.Opus 5 ate the default. Fable is waiting.Claude Opus 5 is Fable-adjacent at $5/$25. That is the default now. Fable 5 still has a job.

New on BenchLM

Ask before a retired model breaks the runRadar Pro covers all 20 providers, five signed webhooks, 100,000 API requests per month, and ten focused views for $9.99/month. The seven-day trial requires a card.

Archive copy reflects the rankings, prices, and availability stated when this issue was sent. Current pages may show newer evidence.

Past issues

Read before you subscribe

Every archived issue is dated. Its claims stay tied to the evidence available on that send date.

OpenAI is keeping ZDR. The safety system is moving.Private Safety Processing is a September preview, not a finished contract.Daybreak Red does not replace BlueGPT-5.6 Cyber vs Sol: the right Daybreak tier, plus persistent Codex Cloud scans.Open weight is eating the frontierDeepSeek shipped 0731 weights; Qwen3.8-Max remained a hosted preview.The frontier is outrunning the scorecardsKimi K3 followed GPT-5.6 Sol by seven days. Comparable evidence is the scarce input.GPT-5.6 arrives, locked behind a previewGPT-5.6 lands in a government-gated preview, Anthropic holds the top two spots, and GLM-5.2 enters the top five.The frontier moved again, twiceOpus 4.7 reaches general availability, Kimi 2.7 enters preview, and the gap between major releases keeps shrinking.

Sponsored editions

Paid partner briefs are labeled before the first partner link. Provider-run claims stay attributed to the sponsor.

The 158 ms call boundarySponsored by Celeris. A bounded look at routing, extraction, scoring, and query rewriting with Celeris-1.
Subscribe to the weekly brief