MyTechDailyNews General: How LLM Price Cuts from OpenAI and Anthropic Reshaped Agent Use in 2026

Pricing announcements for large language models may look like a spreadsheet only developers care about, yet they now decide how much everyday people spend when they ask an assistant to draft slides, review code, or run multi-step research. For MyTechDailyNews General readers, the interesting part is not the math alone; it is the story behind the math: two model makers, OpenAI and Anthropic, repeatedly pushed costs lower while rivals supplied cheaper open-weight options, and the result made long agent sessions less punishing.

Each price string has three positions: input MTok / output MTok / cache read MTok.

What changed in 2026, month by month

Rather than read the announcements as isolated press releases, it helps to see them as a chain. Each move had a practical effect: a cheaper model family, a deeper cache discount, a permanent price, or a forced response to a rival.

Date	Move	What the price did
June 26, 2026	OpenAI introduced GPT-5.6 Sol and Luna.	Sol kept the GPT 5.5 number at $5 / $30 / $0.50. Luna arrived as a smaller option for cheaper jobs at $1 / $6 / $0.10. Terra also appeared between them, but it drew little attention.
June 30, 2026	Anthropic launched Claude Sonnet 5 and placed it nearer Opus, but reaction split.	Opening price: $2 / $10 / $0.20, described at first as temporary and expected to climb later.
July 24, 2026	Anthropic unveiled Claude Opus 5.	This was a major step after Opus 4.5 had affected agentic coding. It cost $5 / $25 / $0.50, near GPT-5.6 Sol. Some benchmarks passed the much larger Fable 5, though real-world response was mixed.
July 30, 2026	OpenAI lowered GPT-5.6 Luna by 80%.	New level: $0.20 / $1.20 / $0.02. The article says the model-quality-to-cost-per-task view implied it would have trailed DeepSeek V4 Pro without this cut.
August 10, 2026	Anthropic made the Sonnet 5 launch discount permanent.	August 10, 2026 turned a temporary introductory rate into the ongoing rate.
August 21, 2026	OpenAI announced a ~20% cut on GPT-5.6 Sol.	Temporarily, GPT-5.6 Sol moved to $4 / $20 / $0.40 for three months.
September 1, 2026	Anthropic announced Fable 5.1.	Price: $10 / $50 / $0.25. It beat Fable 5 from June in every benchmark area and reduced cache read from $10 / $50 / $1 by 75%.
September 3, 2026	OpenAI announced GPT-6 Astra.	It was similar in size to Fable 5.1 and priced at $10 / $50 / $1, but lacked the cheaper cache read. Astra sat close to Fable in performance.
September 22, 2026	OpenAI announced GPT-6 Sol and GPT-6 Luna.	GPT-6 Sol came at $2 / $10 / $0.20, halving input and output relative to the temporarily discounted GPT-5.6 Sol. GPT-6 Luna also came at half prior price: $0.10 / $0.50 / $0.01.
September 22, 2026	Anthropic announced Claude Opus 5.5.	Anthropic explicitly said Opus 5.5 requires less compute to serve than Opus 5, and its pricing reflects that. Its $4 / $20 / $0.20 total was far below Opus 5 because of a 20% input/output cut and a larger cache cut, lowering total cost 40-60% while beating Fable 5.1.
September 29, 2026	OpenAI replaced GPT-6 Sol with GPT-6.1 Sol.	September 29, 2026 was the launch point; after GPT-6 Sol proved underwhelming in performance, a week later GPT-6.1 Sol matched Astra, as personal use confirmed, and OpenAI cut cache price in half to $2 / $10 / $0.10, matching an Anthropic 5%-of-input cache read cost and making the family cumulatively 75% cheaper.

Why cache read pricing changed everyday bills

In coding-agent use, the cache can carry the whole prior conversation on every turn, and that turns cache read into the dominant line item: 96% - 99% of token spend. The hit rate can run around 90 - 100%, depending on inference infrastructure. When a cut removes 75% of cache read cost, the whole bill can drop by 70-75%.

What the wider market showed

From the GPT-5.6 launches in June to now, popular LLM usage cost has cratered; anyone looking from a year ago would find the scale hard to believe. Paid harness subscriptions such as Codex and Claude Code also benefit. The article says output quality does not appear to have been sacrificed for lower prices. Vercel AI gateway data supports cost-per-token falling to 43.3% of the April 2026 cost, while total spend rose because model improvement kept outpacing price reduction.

Why prices behaved like commodity markets

When people test agents with each new LLM, subjective vibes are often the only differentiator. Since last November triggered hypergrowth of agentic LLMs, models behave more like commodities because switching is easy through OpenCode or Zed Agent, or through a changed API endpoint string on OpenRouter. That leaves little moat. There are many competitors beyond Google and Meta, offering enough substitutes to create pricing pressure.

Open-weight models, provider routing, and the 2026 supply picture

DeepSeek Flash V4.1, including its updated 1017 variant, shows the effect. On OpenRouter, as retrieved October 24th, 2026, the original DeepSeek V4 Flash 0420 appeared in April 2026 at $0.14 / $0.28 / $0.0028 and went megaviral partly because OpenAI lacked a simple-task small model. The key was $0.0028: a cache read proportion of 2% is far below 5%-10% for OpenAI and Anthropic models, making effective cost lower than the headline suggests. The article suspected that the July 80% GPT-5.6 Luna cut targeted these smaller open-weight models. DeepSeek Flash V4.1 1017 now scores significantly better than DeepSeek Flash V4 with 45 on the Artificial Analysis index versus 24, while DeepSeek still charges $0.15 / $0.60 / $0.003. Note the operative words from DeepSeek.

Any inference provider can list a hosted model on OpenRouter, and requests through the OpenRouter API go to available providers by default or by preferences such as lowest price. Back in May, there were 13 distinct providers hosting DeepSeek V4 Flash. For the updated variant, the sorted provider list for DeepSeek V4.1 Flash 1017 shows 29 distinct providers; the list was retrieved October 24th, 2026. When effective input costs are weighed with cache read and hit rates, many providers are close. Cloudflare and DeepInfra were within 10% of DeepSeek cost, though these cheapest models still held only a minority of total token share.

OpenAI and Anthropic need to differentiate from open-weight alternatives with higher-quality models; they are doing so, especially with Opus 5.5, but must keep doing that, which is why billions of dollars are going into R&D. Open-weight models are catching up: just a few days ago, Moonshot AI announced Kimi K3.1 Pro, comparable to Opus 5.5 with Kimi at 55 versus Opus 5.5 at 58, at half input/output price and same cache read: $2 / $10 / $0.20. It shares Kimi K3 size and architecture; Kimi K3 is hosted by over 20 inference providers. Once weights drop on November 4th, most can be expected to support the new model at comparable prices. Moonshot AI also joined the price cuts, pricing Kimi K3.1 Pro below the original Kimi K3 at $3 / $15 / $0.30.

Durable low prices instead of a later snapback

Critics often claim tokens are priced low only to hook users and will later rise to the true cost of a token, wrecking boosters. The analogy is Uber and Lyft in the 2010s, but unlike that duopoly era, OpenAI and Anthropic cannot set prices alone and inference entry barriers are lower. If many inference providers offer similar quality at similar prices and switching costs are negligible, prices may keep falling more. That is why both companies seek moats such as enterprise commitments and personal agentic assistants like OpenAI Dots and Claude Duet.

A sudden jump in token prices conflicts with behavioral economics and the observed significant declines for better models, not only from OpenAI and Anthropic. As long as scalers charge below revenue per token, providers will reduce costs to compete until OpenAI and Anthropic follow. The 29 distinct providers on OpenRouter have little reason to serve an unprofitable model; they would stop and redirect capacity, making the true cost of a token possibly equal to marginal cost.

How IPO pressure enters the price logic

Both OpenAI and Anthropic are close to IPOing, raising questions about price cuts. OpenAI is rumored to IPO in 2027, and Anthropic within a couple months. Leaked financials show they are losing billions from R&D/scaling, though reports do not capture recent agentic expansion. Investors will examine negative profits and demand plans.

Why cut a primary revenue source by up to 80%? That seems opposite to looking attractive in an S-1 prospectus. Yet inference profit margins may differ; some reports say profit-per-token margins are as healthy as 80% with caveats. How much margin remains after cuts?

The article describes one user response that may not match lab expectations. Before GPT-6.1 Sol, the writer had a $100/month Codex subscription, the 5x Pro plan, while GPT-5.6 Sol was the top model, and exhausted all included usage despite annoying random resets. After GPT-6.1 Sol launched and effective usage cost dropped by 75%, the writer cut the max usage quota to 1/5th because the same model was 4x more efficient, roughly neutral.

Subscription margins for Claude Code and Codex remain uncertain because their value may exceed API usage; if capital tightens, the article suspects allowances would be cut before base prices. At the same GPT-6.1 Sol announcement, OpenAI reduced the usage allowance by 50% for the $200/month tier in Codex under the argument to reflect our increasingly efficient models, technically correct, though funny to put in writing.

Anthropic IPO prospectus and public inference margins are also part of what comes next.

Obsolescence, infrastructure, and the next releases

It may not benefit OpenAI or Anthropic to make pricier models obsolete intentionally. After Opus 5.5 released a month ago, there has been little discussion of Fable 5.1 or GPT-6 Astra because Opus 5.5 is just that good. A top-and-left view from Anthropic reinforced that position. The writer expects Fable 5.2 or GPT-6.1 Astra to release in a couple weeks and prove them wrong, and will be curious how it is priced.

A less cynical explanation is that LLM infrastructure has improved. The article believes Anthropic when it says Opus 5.5 requires less compute to serve than Opus 5, and its pricing reflects that. If efficiency reduces need for massive data centers, that could cause cascading economic failures, but likely not, because inference is only one part of compute mix and data centers remain needed for the pipeline.

Checklist for readers deciding what to watch

• Look at the full price string, not only input: input MTok / output MTok / cache read MTok.
• Remember cache read can dominate agent spend, especially at 96% - 99% of tokens in coding use.
• Check whether a cut is temporary or permanent; the Sonnet 5 launch discount became permanent on August 10, 2026.
• Treat open-weight options as a market check: DeepSeek Flash V4.1 and Kimi K3.1 Pro both show how fast alternatives can move.
• Watch subscriptions separately from API prices, because Codex and Claude Code may behave differently under margin pressure.

Common misconceptions to avoid

1. Assuming a headline price tells the whole bill. Cache read and hit rates can change the result by 70-75%.
2. Assuming all cuts are a short-term promotion. The source argues low prices may last because switching is easy and providers compete.
3. Assuming lower prices mean lower quality. The article says output quality does not appear to have been compromised.

FAQ

What does the 2026 price timeline show?

It shows a sequence of competitive price changes involving OpenAI, Anthropic, DeepSeek Flash V4.1, and Moonshot AI, with cache read becoming a central lever.

Why is cache read such a big deal for agents?

Because the cache can hold the whole prior conversation, it can be 96% - 99% of token spend, with hit rates around 90 - 100%.

Does provider diversity really lower my cost?

The article says many providers can list models and route requests by preferences such as lowest price. That competition can narrow effective cost, as with Cloudflare and DeepInfra near DeepSeek cost.

Action advice for general users

1. Before moving an agent workload, compare total cost per completed task, not one number from the launch post.
2. If cache-heavy work is involved, look first for cache read reductions and hit-rate assumptions.
3. Keep an eye on whether discount language says temporary, three months, or permanent.
4. Treat new IPO and margin disclosures as useful signals, but judge value by the model you actually use.

Final read

For everyday users, the odd thing about 2026 was not that prices fell, but that the market kept rewarding faster service, cheaper inference, and wider provider choice while still delivering better models. Even if Anthropic and OpenAI IPOs turn out badly enough to reshape those companies, the source argues that LLMs and agents as an industry will not die suddenly, because open-weight models and secondary providers continue to act as an economic check on pricing and other lab behavior.