Every's 9-Person Team Spent a Month With It: How Good Is GPT-5.6 Sol, Really?

One Month Without It, and the Team Says It Felt Like the Stone Age

Every is a US-based AI media company that also builds AI products — publishing deep analysis while shipping its own tools. Its editorial team regularly puts frontier models through hands-on testing, and this "vibe check" series is their signature format. By the time this review went live, Sol had already been living inside Every's daily workflow for about a month — deeply enough that losing it actually hurt.

During that month, Sol was everywhere. It helped Dan Shipper keep his inbox at zero, tracking decisions scattered across meetings and Slack that he'd otherwise have missed. It kept pace with Austin Tedesco — from a marketing idea, to an email, to a landing page, to an experiment — without him ever having to repeat himself or lose his train of thought. For this piece's author, Katie Parrott, it fetched files and pulled context so fast it completely rewired how she worked with models day to day.

In late June, Sol went offline for a round of government review, and the team lost access. Dan said going back to other models — even with Fable still on hand — felt like regressing to the Stone Age. Austin compared switching to GPT-5.5 to "shooting a basketball that's twice as heavy as usual." Every had earlier floated an idea in its Sonnet 5 review — "a revolution of ever-rising expectations" — and this Sol-less stretch proved exactly that: how fast people adapt to a higher standard of living, and how painful it is to slide back.

Dan has a sharper analogy: Sol is a Porsche, Fable is a warp drive. Fable can obviously carry you across an entire galaxy, but most of the time you're not headed to space — you just want to get around town, and Sol is the car that makes that drive feel good.

Nine People, Nine Honest Takes

Every's "Reach Test" has nine team members, each in a different role, give one honest line on Sol along with a sentiment rating. At a glance, it's clear this reliance isn't just for show.

Eight of the nine cards lean positive; one (Arielle) sits at "fine" after a late-stage calculation error. That overall lean explains why the team is willing to treat it as a daily driver. What it's actually good at, and where it falls short, gets broken down section by section below — coding, writing, and knowledge work.

It Can Carry the Load — It Just Doesn't Know When to Stop

In day-to-day development, Sol is a genuine upgrade over GPT-5.5: it can chase a bug through an unfamiliar production codebase, carry a large project from start to finish, and keep testing well past where other models would give up. Several engineers on the team have already made it their daily driver. Its limit shows up on tasks that require the model to judge, on its own, what it shouldn't do.

Every has a benchmark called Senior Engineer: hand the model a real but messy collaborative codebase and ask it to judge, like a senior engineer would, what the system should be rewritten into — specifically testing whether it knows when to stop and avoid over-engineering. On this benchmark, Sol scored 56; Claude Fable 5 scored 90.

Where the Goal Is Clear, It Shines

On the flip side, when the target is well-defined, Sol's execution holds up. The strongest production case came from Naveen Naidu, who used Sol in daily development on his own product, Monologue: GPT-5.5, even at its highest reasoning setting, failed repeatedly to root-cause a note-recording bug; Sol traced the failure through the existing codebase and fixed it. Kieran Klaassen also had Sol and Fable rebuild the team's collaborative document editor, Proof, from a single prompt — Sol delivered a working Proof-like app in roughly a third of the time (though Dan preferred Fable's version on design). Sol also finished a digital audio workstation that GPT-5.5 hadn't been able to complete.

These results line up with the team's overall impression: once the desired system is clear, Sol is fast, resourceful, and genuinely capable at implementation. When "judging what not to build" is the main engineering task, Fable remains the first choice; once direction is set, handing execution to Sol is usually the path of least friction.

Last Place in the Writing Blind Test, Yet the Team's Most Indispensable Partner

In Every's writing benchmark, Sol ranked last of six models — its prose was the hardest to read, and its editorial choices strayed furthest from the published reference version. And yet the whole team overwhelmingly prefers using it for daily writing, over both Sonnet 5 and Opus 4.8. The answer lies in a framework that runs through the entire piece.

All four scenarios tested — coding, writing, knowledge work, and agent tasks — can be plotted on this same axis: lean left (independent judgment calls, one-shot then review) and Fable is more reliable; lean right (someone watching and adjusting in real time) and Sol works better. Most everyday writing, it turns out, sits toward that right side.

This review itself is a case in point: Sol could track the tonal arc of dozens of past Vibe Check pieces, dig up an old Slack conversation buried somewhere that supported a specific detail, and push out 24 drafts in one 6-to-8-hour focused session — without the author ever having to sit and wait for it to slowly work things out. Every draft was a string of editorial calls, each one nudging the piece closer to what it should be. For writing — a task that's inherently iterative and collaborative by nature — that's what a strong model should look like, even if nailing an opening line in one shot, or sticking a landing the way a human would, is still out of reach.

Why It Bombs the Blind Test

The writing benchmark is deliberately narrow — it only measures how the model performs unassisted, with no help from a human. Sol came in last: lowest similarity to the published version, and the highest Flesch-Kincaid score, meaning the lowest readability. Flesch-Kincaid (FK) is a formula that scores text based on sentence length and word difficulty — the higher the score, the harder it is to read.

Every's writing benchmark: similarity, grade level, and readability compared across six recent models. GPT-5.6 Sol ranked last — lowest similarity to the published piece, highest FK score, lowest readability. Image / Katie Parrott, Every

Shorter Sentences, Bigger Words

Sol tends to write relatively short sentences, breaking an argument into tight little paragraphs. In its strongest opening draft for "After Automation," its rhythm tracked closer to the published opening than Opus 4.8's did, while Opus built out fewer, larger paragraphs. The problem shows up inside the sentence itself: even when both models produce sentences of similar length, Sol favors longer, more abstract words.

Opus uses plainer words inside more complex syntax; Sol delivers text that scans smoothly at a glance but reads more laboriously line by line. Visible AI tells, oddly enough, aren't what drags it to last place: on checks for stock transitions, false contrasts, clichés, and repetitive rhetorical patterns, Sol lands in the middle of the pack. Its prose can be clean — just not always clean enough that you can skip re-checking for that machine-y tone before hitting publish.

GPT-5.6 Sol writing sample: leans toward short sentences and tight paragraphs, with fewer visible AI tells. Image / Katie

Opus 4.8 writing sample: more complex syntax, longer paragraphs, more prone to falling into "not X, but Y" constructions. Image / Katie

Give It Context, and the Quality Jumps

The blind test deliberately withholds material, but in real use, people feed models plenty of it. Once you see how Sol absorbs and uses that context, its stronger side comes out. The author gave Sol and Opus 4.8 the same writing task, the same source material, and the same direction, and had each write a draft opening for the column "Working Overtime." Sol turned in a hook that grabbed attention, and the whole piece tracked closer to the column's established tone and voice; Opus turned in dense, hard-to-read paragraphs that didn't match the voice the author had defined in her context file.

Sol (with the author's material and style context): faster, and closer to a draft she'd actually take and edit. Image / Katie

Opus 4.8 (with the same material and context): dense, complex sentences and ornate word choices, a mismatch for the author's more conversational tone. Image / Katie

Austin sees the same pattern with marketing copy: with no guidance, Sol is "a writer anyone could be," generic and repetitive; once given company background, templates, and a style guide, the landing-page copy, social posts, and marketing emails it produces need so little editing that Austin will just ship them. Give it material, examples, and rules, and quality climbs; let it decide the argument and the bar on its own, and the weaknesses from the blind test come right back.

As a Live Writing Partner, It's at Its Best

Sol's strongest role is as a real-time writing partner. It revises fast and stays aligned with editorial direction, letting a writer try a new opening, reshuffle a paragraph, or scrap a weak draft entirely without having to re-brief the task each time. For the author's own interview-heavy, iterative writing process — full of directional feedback and backed by a style doc and examples — Sol clearly beats the Claude lineup: Opus responds slowly, Sonnet 5 is hard to steer, while Sol switches direction fast, fixes mistakes fast from feedback, and carries new information forward into later turns instead of fixating on the most recent instruction and losing the bigger goal the way Sonnet 5 does. Head of technical consulting Mike Taylor has also started handing writing and editing to Sol, because it "basically never says anything off-putting" — it takes correction, remembers sources and prior decisions, and quickly tries again. Judgment still comes from a human editor, but Sol makes that judgment cheaper to apply across an entire piece.

Same Task, Three Models, Three Different Approaches

Saying abstractly that "Sol takes initiative and asks questions" isn't as useful as one real example. Every's head of operations, Arielle Shipper, gave GPT-5.5, Claude Fable 5, and Sol the exact same starting task and watched how each one handled it.

Austin's own daily workflow shows the same coherence: he can start from a marketing idea, draft an email, turn the copy into a landing page, and set up an experiment — all without ever leaving Codex or re-explaining the audience and pitch. Senior editor Jack Cheng uses Sol to merge paragraphs and strip out jargon, doing it right on the public page where the copy will actually appear, so every line can be judged in its final context. These workflows look less like a one-shot app and more like an actual workday: find the source, understand the need, ask about the decisions that need a human call, and carry the answers into the finished product.

Three Pricing Tiers, Two Reasoning Modes

On pricing, OpenAI's three tiers line up almost one-to-one against Anthropic's three models, making cross-vendor budget comparisons easy.

Side by side: Sol matches Opus 4.8 on input price and costs $5 more per million output tokens; Terra is pricier than Sonnet 5's introductory promo rate ($2 / $10, expiring August 31) but close to the post-hike rate of $3 / $15; Luna matches Haiku 4.5 on input and costs $1 more on output.

What max and ultra Are

Beyond the three model tiers, Sol also brings two new reasoning settings, letting you scale compute spend to task difficulty instead of running everything at the same setting.

On the product side, Every hasn't used the merged ChatGPT-and-Codex app much yet, but the early feel matches what Codex was like at launch: a place worth sticking around in.

When to Reach for Sol, When to Switch to Fable

Pulling together the judgments from the four scenarios above into a checklist you can actually use. The original piece gives two sets of scenarios, mapping neatly onto the two ends of the delegation-vs-collaboration axis.

Every's team usage is drifting toward this same axis: Dan hands hard engineering work to Fable as the lead agent, and well-defined execution to Sol; Mike treats Sol as his daily driver, saying it has "finally displaced Opus," while still keeping Fable around for its "sharp peak intelligence" and stronger use of context. When you want a clear outcome and an agent that keeps pushing forward while you stay free to jump in anytime, use Sol; when "defining the system itself" is the main task, either set checkpoints early and often, or hand the architecture to Fable and let Sol handle execution.