← The Polylog AI Intelligence Brief
Morning Edition · Tuesday, October 27, 2026Published at 2:18 AM EDT · New York
OpenAI prices GPT-6.1 Terra at $1 and $5 per million tokens, half the rate of GPT-6.1 Sol
OpenAI claims the mid-tier agent model matches GPT-6 Astra on Agents' Last Exam at a tenth of the token price and scores 66.7% on OSWorld 2.1 at medium reasoning effort.
OpenAI released GPT-6.1 Terra on October 26 for agentic work, computer use and everyday coding. The model is available through the API as gpt-6.1-terra. Standard pricing is $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. Batch and Flex run at half that rate and Fast mode at double, according to developer coverage. A Russian-language summary frames the release as a halving of the price of OpenAI's mid-class agent model, with input falling from GPT-5.6 Terra's $2. The context window is 1.1 million tokens.
OpenAI's main capability claims cover computer use and agents. On the OSWorld 2.1 offline set, Terra scores 66.7% at medium reasoning effort and 74.2% at maximum, The Decoder reported. OpenAI says the medium-effort score already exceeds GPT-5.6 Terra's best result. The company also says Terra matches GPT-6 Astra on Agents' Last Exam at one-tenth of Astra's per-token price and finishes tasks more than three times faster than Astra and GPT-6.1 Sol. All of these numbers are vendor-reported. A write-up by BERI citing Artificial Analysis puts Terra four points below Sol on that firm's index, which is roughly what the price gap would suggest.
The cut follows a pattern. On July 30, OpenAI cut GPT-5.6 Terra's prices by 20%, to $2 per million input tokens and $12 per million output tokens, and cut GPT-5.6 Luna's price by 80%. At the time it cited a 20% reduction in end-to-end serving costs and better speculative decoding, a technique that speeds up token generation. Meta's Muse Spark 1.1, launched in July, sits at $1.25 per million input tokens and $4.25 per million output tokens. Terra's input price is now below Meta's and its output price is close to it.
Agent builders benefit first. Computer-use loops consume large volumes of input tokens from screenshots and long context, so input price drives per-task cost. Code-review companies already running OpenAI models, such as Greptile with GPT-6 Astra, now have a cheaper tier to route routine checks through.
What this means
The mechanism is per-task cost in agent loops, where input tokens dominate and a halved input price changes which workloads are worth automating. Application companies that resell model calls gain margin. Anthropic is the most exposed lab, because its filing on the same day shows that the 80% of its revenue coming from metered consumption can shift between providers quickly. OpenAI can afford the cut because it attributes earlier reductions to measured serving-cost gains, so this is partly a pass-through of efficiency rather than a pure subsidy. The market already expects falling token prices. The new element is the vendor claim that a mid-tier model matches the flagship on an agent benchmark, which would sharply reduce demand for the flagship tier on routine work if independent tests confirm it. The precedent is OpenAI's July 30 cut, which came within weeks of launch, and Meta's July entry at aggressive prices. The desk's read is that Anthropic, Google or Meta reprices a mid-tier model within weeks, provided their serving costs have fallen at a similar pace. The read is wrong if rivals hold prices and Terra's share of routed traffic on aggregators stays flat, which would mean buyers are choosing on capability rather than price.
Part of a tracked trend
Frontier Model Price War
Frontier API vendors increasingly compete on strategic price-cutting rather than pure capability, repeatedly launching or repricing models below prevailing rates to grab share and compressing industry inference margins; expect recurring below-rival pricing moves.
Townsquare
Start a discussion in Townsquare.
More from this edition
- OpenAI says its internal models refuted the Exponential Time Hypothesis and proved the Kakeya conjecture in every dimension
- Anthropic files to list on Nasdaq as ANT, disclosing $16.3 billion in first-half revenue against hundreds of billions in compute commitments
- A two-neuron steering recipe removes most refusals from Alibaba's Qwen3.5-4B, and an independent replication confirms it
- Anthropic reports that Claude ran a 64-hour autonomous protein-design campaign with 17 lab-confirmed binders
- Kevin Weil raises $215 million for Warp Sciences to build automated drug-chemistry labs in the United States
- Allen Institute's FairStress benchmark finds models abandon correct decisions more readily when the correct answer favors a minority-coded candidate
- Berkeley Lab runs Meta's open-weight vision models on 300 graphics processors to cut beamline image analysis from a month to 15 minutes
Comments
0No comments yet.