Updated October 29, 2026: rebuilt on the September–October model releases. The previous version, written against the June–July lineup, said Claude was 17% cheaper on output at the two middle tiers and that GPT-5.6 Luna had no Claude equivalent. Neither is true any more; the analysis below replaces it.
1. What changed since August — the second lineup reset
Our August version of this page described two ladders separated by cents, with Claude slightly cheaper in the middle and OpenAI far cheaper at the bottom. Eight weeks later that picture has inverted. Anthropic shipped four models in five weeks — Fable 5.1 (September 1), Opus 5.5 (September 22), Sonnet 5.5 (September 28) and Haiku 5.5 (October 7) — while OpenAI moved to the GPT-6 and GPT-6.1 families. The bottom of the market, where OpenAI had the field to itself, is now priced identically. The middle, where Claude had a modest edge, now belongs to OpenAI at roughly half the price. “Claude or GPT?” is still the wrong question; the routing table that answers the right one has changed.
2. Side-by-side
| Tier | Claude (Anthropic) | GPT (OpenAI) |
|---|---|---|
| Frontier | Fable 5.1 — $10 / $50 per M tokens (Sep 1, 2026) | GPT-6 Astra — $10 / $50 |
| Workhorse | Opus 5.5 — $4 / $20 (Sep 22, 2026) | GPT-6.1 Sol — $2 / $10 |
| Mid-tier / agents | Sonnet 5.5 — $2 / $10 (Sep 28, 2026) | GPT-6.1 Terra — $1 / $5 |
| Volume / classification | Haiku 5.5 — $0.10 / $0.50 for prompts up to 100k tokens; $0.50 / $2.50 above (Oct 7, 2026) | GPT-6 Luna — $0.10 / $0.50 |
| Last round | $65B Series H (May 28, 2026) | $122B round (Mar 31, 2026) |
| Valuation | $965B | $852B |
| Total raised | ~$132B; confidential IPO S-1 filed Jun 1, 2026 | ~$180B |
| App-layer posture | Substrate only — no competing GTM application | Workspace Agents into Slack, Salesforce, Drive (Apr 22, 2026) |
| Best-fit GTM use | Long-horizon agents where your own evals favour Claude; volume work with short prompts | Cost-sensitive mid-tier agents, broad seller copilots, volume work with long prompts |
API list prices per million input / output tokens, from each vendor’s own pricing page in October 2026. The tier pairing is our inference — neither vendor maps its models to the other’s — and it matches models by position on each ladder, not by measured capability.
3. Identical at the ends, half-price in the middle
Read the ladders tier by tier and the pattern is stark. At the frontier, Fable 5.1 and GPT-6 Astra both list at $10 / $50. At the volume tier, Haiku 5.5 and GPT-6 Luna both list at $0.10 / $0.50 — with one qualification that matters for GTM: Haiku’s price rises five-fold, to $0.50 / $2.50, once a prompt exceeds 100k tokens, and Luna’s published rate carries no such step. Short classification and routing prompts sit comfortably under that line; a whole call transcript plus CRM history may not.
The two middle tiers are where the vendors now diverge, and they diverge the opposite way from August. GPT-6.1 Sol at $2 / $10 is half the price of Opus 5.5 at $4 / $20, and GPT-6.1 Terra at $1 / $5 is half the price of Sonnet 5.5 at $2 / $10. These are the tiers most GTM agent work actually runs on — CRM-writing agents, account research, RevOps automation — so a 2x gap on list price is the number that compounds on the invoice. Anthropic did cut Opus by 20% (from $5 / $25 under Opus 5) and held Sonnet flat; OpenAI simply cut further.
List price is not cost per task. A model that finishes an agentic job in fewer turns, or with fewer retries, can be cheaper at twice the rate — which is why the only honest way to settle the middle tier is to run your own workloads on both and price the completed task, not the token. What has changed is the burden of proof: in August, Claude’s middle tiers were cheaper and you tested GPT to see if it was good enough; now GPT’s are cheaper and Claude has to earn the premium on your evals.
4. Availability is now a selection criterion
The most durable lesson of Q3 was not a benchmark. On June 12 a US export-control order forced Anthropic to suspend global access to Fable 5; the order was lifted June 30 and the model returned worldwide on July 1. We covered the episode in the Q3 capability tracker, and the operational read has not changed: frontier-model availability is now a regulatory variable, not just a reliability one. Eighteen days is longer than most GTM teams’ tolerance for a dead enrichment waterfall or a research agent that stops answering.
This cuts against both vendors, and it is the strongest argument in this entire comparison for refusing to pick one. Teams that had pinned a single model into a single code path spent June rewriting under pressure. Teams that had built model-agnostic routing — the pattern in our agent-native playbook — changed a config value. With the MCP 2026-07-28 specification now final, the plumbing for that portability is standard rather than bespoke, which removes the last good excuse for hard-coding a provider.
5. The channel conflict has not gone away
One structural asymmetry survives the lineup reset intact. OpenAI’s Workspace Agents, shipped April 22, plug directly into Slack, Salesforce and Drive — the exact systems a GTM tool integrates with. If you are building an SDR or RevOps agent on GPT, your substrate supplier now ships a product into your category. Anthropic, so far, does not: it sells the model and the developer surface around it, and has not shipped a GTM application competing with what its customers build. (Claudeforce, announced at Dreamforce in September and in beta, is Salesforce’s product built on Claude — a partner channel, not an Anthropic application.) That is a genuine difference in supplier risk, and it matters most to the buyer who is building something durable on top rather than buying a copilot for this quarter.
The honest counterweight is distribution, and it is just as real. GPT is the default model inside ChatGPT, which means it is already on your sellers’ phones and already the thing your marketers paste into without filing a procurement ticket. Ambient adoption is a moat no benchmark captures, and OpenAI’s July price cuts read as a company defending exactly that position — buying the volume tier outright rather than arguing about frontier scores.
6. Verdict — re-run the routing table
The portfolio approach survives; the defaults inside it change. In the middle tiers, start from GPT-6.1 Sol and Terra and make Opus 5.5 and Sonnet 5.5 earn their 2x list premium on your own completed-task costs — they may well do so on long-horizon agent work, but that is now something to measure rather than assume. At the volume tier, the choice is a wash on price: pick on quality for your task, and keep prompts under Haiku’s 100k-token step if you run on Claude. At the frontier, price no longer separates them at all, so supplier risk does — and the channel-conflict argument above still points the same way. Where sellers already live in ChatGPT, let them.
Above all of it sits the rule June made non-negotiable and October reinforced: build on the substrate you can swap. Two full repricings in a single quarter is the clearest evidence yet that today’s cheapest tier is a fact with a shelf life. Route through an abstraction, keep a tested fallback at every tier, and price your architecture on completed tasks rather than launch-day list prices. As our Build-vs-Buy analysis put it in a different context — spend frontier tokens on judgment, commodity tokens on everything else. The teams that could move between ladders in an afternoon captured both resets; the teams that hard-coded a provider captured neither.