The 2026 SaaS pricing divide: retreat or barbell?

Over one week in September 2026 our daily-refresh agent caught fifteen SaaS vendors mid-repricing. Not gentle 6% price bumps. Structural rewrites — tiers renamed, floors doubled, free plans killed, whole ladders rebuilt.

When we sorted the fifteen by what direction they moved, they fell cleanly into two opposite camps.

Half of them retreated upmarket. The other half barbelled — free at the bottom, $200+ at the top, and a $20 middle that used to be $50.

Both are responses to the same problem: seat-based SaaS pricing has broken under AI. This is how vendors are choosing to survive it.

The Retreat playbook

The retreat move has three steps: kill the free tier, raise the paid entry, quietly deprecate everything under a mid-market floor.

Tabnine did all three. Free and individual plans were retired in April 2025. The floor is now $39/user/month for Code Assistant, $59/user/month for the Agentic Platform, both annual-billing only. No monthly option, no self-serve free tier, 14-day trial replacing what used to be a permanent free plan. The old $9 individual plan is dead.

Sourcegraph Cody went further. On July 23, 2025, Sourcegraph discontinued Cody Free, Cody Pro, and Enterprise Starter — all in one move. Cody is now available only on the Enterprise plan at $59/user/month with a ~$16,000 team minimum. Solo developers, indie hackers, freelancers, and small teams are simply not the customer any more.

Common Room repriced the same week Zoom's acquisition closed. Old ladder: Free / Team $625 / Business $1,600. New ladder: Essential $2,500/month (billed annually) / Advanced custom / Enterprise custom. The free plan is gone. Monthly billing is gone. Entry moved 4x.

Krisp killed its free-forever plan and replaced it with a 7-day trial. The paid tiers were renamed Pro→Core and Business→Advanced, and the annual discount deepened from ~25% to ~50%, which quietly makes monthly billing 2x more expensive than it looks.

Same move, four times: give up the individual segment to protect margin at the enterprise segment.

The Barbell playbook

The barbell move is the opposite. Free at the bottom, expensive premium at the top, and a $20 middle where everything used to be $50.

Devin ran the cleanest version of this play. The old ladder was Core ($20 + $2.25 per compute unit) and Team ($500/seat/month, 5-seat minimum). In 2026 Cognition retired both. The new ladder is Free (light quota), Pro $20/month, Max $200/month, plus Teams and Enterprise. Entry moved from $500 to $20 — nominally a 96% cut. In reality, a full repackaging: the seat-based enterprise product got deleted and replaced with an individual quota product.

Windsurf (formerly Codeium, and possibly rebranding again as Devin Desktop under Cognition) restructured on March 19, 2026. Free with unlimited autocomplete, Pro at $20/month, Max at $200/month, Teams at $40/user. Pro moved from $15 to $20 to match Cursor Pro exactly. The credit pool was replaced with daily and weekly refreshing quotas.

Consensus doubled its Pro entry price from $10 to $20 monthly, but added a new $65/month Deep tier for heavy users doing 200+ literature reviews. The middle stayed flat. The premium tier is where they're now capturing the value.

Cursor has been running the barbell for a year already: Free / $20 Pro / expensive Business tier for teams. Every AI-coding competitor is now converging on that exact shape.

Same move, four times: capture a huge free-tier funnel, price the middle at $20 (where AI usage is subsidized by underuse), monetize the top-1% of heavy users at $200+ where their real compute cost gets recovered.

Why both playbooks now

The forcing function is the same: seat-based SaaS pricing does not work when AI usage varies 100x per user.

A single Devin power user can burn through more compute in a day than a hundred casual users combined. If you charge them the same $20, the power user is unprofitable and the casual user is being overcharged. If you charge them the same $500, the casual user churns immediately. There is no logical middle price.

The Retreat vendors solve this by picking one customer: the enterprise. Give up the variable-use middle. Sell only where the seat count is high enough and the deployment is contained enough that per-user usage is predictable.

The Barbell vendors solve this by picking both extremes: $0 for the casual user (who costs almost nothing to serve), $200 for the power user (whose cost you can recover), and a marketing $20 tier in the middle that most subscribers under-use enough to subsidize the ones who don't.

Frontier lab pricing is the second force. Anthropic and OpenAI keep raising API rates. Any SaaS vendor whose margin depends on cheap frontier tokens is watching that margin get compressed quarter-over-quarter. Both playbooks are ways to get out from under that trap: enterprise contracts either bake in the LLM cost, or the barbell's $200 tier passes it directly to the user.

What this means for your 2027 renewals

Three implications, all specific enough to change how you negotiate this cycle.

If your vendor is retreating, your renewal quote will hurt more than list price implies. Tabnine, Common Room, and Sourcegraph Cody customers on grandfathered pricing are being migrated onto the new floor. Ask your rep directly: what's the migration path for my current tier? If the answer is "you'll be moved to Essential/Enterprise pricing", model the full-year impact before you re-sign.

If your vendor is barbelling, you're either on the subsidized side or the subsidizing side. The $20 middle tier is a deliberate loss-leader for the heavy users on the $200 tier. If your team is on Pro and using it heavily, expect that plan to either raise price or add usage caps within 12-18 months. Barbell pricing is not stable at the middle.

Usage-based buy-back clauses are the new must-have. Ask for a contract term that lets you re-price mid-term if your actual usage stays under X% of the licensed capacity. Retreat vendors will resist this hardest — because their model depends on you not asking. Barbell vendors will offer it because usage variance is already priced in.

Every price cited in this post links to that tool's live pricing history on SaaS Price Hub. Our daily refresh agent verifies 30 vendor prices per day; the full 220-tool directory recycles every 7-8 days. See /tracker for the full change log.