Hello, AI enthusiasts! Editor Z here from AITech blogZ. We first ran a version of this comparison back in February 2025, when GPT-4o and Claude 3.5 Sonnet were the names everyone argued about. Almost none of that lineup is still current. Every model mentioned in the original post has been superseded at least once, one company changed its entire naming scheme, and the cheapest serious option on the market now costs a fraction of what it did then. Here is where the five major LLM families actually stand as of September 2026, with the numbers to back it up.
What actually changed since early 2025
The headline shift is that GPT-4o is gone from ChatGPT. OpenAI retired it, along with GPT-4.1 and o4-mini, from the consumer app on February 13, 2026, after usage had fallen to roughly 0.1% of daily chats; it lingers only in the API for developers with existing integrations. Anthropic renamed its whole lineup — "Claude 3.5 Sonnet" became "Claude Sonnet 5," with the version number now following the tier name (Haiku, Sonnet, Opus) rather than the other way around. Google moved from Gemini 1.5 through the entire Gemini 2.x generation to Gemini 3, then Gemini 3.1. DeepSeek replaced R1 with DeepSeek V4 in April 2026. And Grok, which the original post described as available "exclusively through X," now has its own standalone app and API at grok.com, independent of a Twitter/X account.
The other trend worth flagging: prices are falling even as capability rises. Anthropic's newest model, Claude Opus 5.5, is cheaper per token than the Opus 5 it replaced. Google cut its top-tier Gemini subscription from $250 to $200 a month. DeepSeek's open-weight models remain 10-30x cheaper per token than their closed-source rivals. If you compared prices even six months ago, they're already out of date.
The current lineup, side by side
This table compares each company's current flagship-tier model — the one most likely to show up as the default paid option in its respective app — rather than every size variant each company sells.
| Model | Released | Context window | API price (input / output per 1M tokens) | Best for |
|---|---|---|---|---|
| GPT-6 Astra (OpenAI) | Sept 3, 2026 | 1.05M tokens | $10 / $50 | Frontier reasoning, long-document and agentic/computer-use work |
| Claude Opus 5.5 (Anthropic) | 2026 | Up to 1M tokens | $4 / $20 | Coding, long-context reasoning, enterprise use |
| Gemini 3.1 Pro (Google) | Feb 19, 2026 | 1M tokens | ~$2 / $12 (Gemini 3 Pro-tier pricing) | Multimodal tasks, huge documents, tight Google Workspace integration |
| DeepSeek V4-Pro | April 24, 2026 | 1M tokens | $0.66 / $1.98 (off-peak) | Coding and math on a budget; open-weight self-hosting |
| Grok 4.7 (xAI) | 2026 | 500K tokens | $2.20 / $6.60 (under 200K-token prompts) | Real-time/X-integrated queries, agentic browsing |
A reader who only has time for one number: DeepSeek is the budget option by a wide margin, Gemini and Claude's mid tier (Sonnet 5, at $2/$10 per million tokens) are the best value for general use, and GPT-6 Astra and Claude Opus 5.5 sit at the top of the price range for the hardest problems.
OpenAI: GPT-6 Astra, and what actually runs by default
OpenAI's newest model, GPT-6 Astra, launched September 3, 2026, and is being rolled out in phases — companies in OpenAI's cybersecurity-focused access program got it first, given its strength in computer-use and agentic tasks (OpenAI reports it completes OSWorld 2.0 benchmark tasks in 47% less time than the prior model). But Astra is not what most ChatGPT users see day to day. As of mid-September 2026, everyday chats on paid ChatGPT plans run on GPT-5.6 Sol, with free and Go-tier users on the lighter GPT-5.6 Luna. Subscriptions run Free, Go ($8/month), Plus ($20), Pro ($100), and Pro Max ($200) — a wider ladder than the original 2025 post's simple ChatGPT/API split.
Anthropic: a renamed lineup, and prices moving down
Claude's naming convention changed entirely. What would once have been "Claude 4 Sonnet" is now Claude Sonnet 5, priced at $2 per million input tokens and $10 per million output tokens — a rate Anthropic had originally planned to raise to $3/$15 in September 2026 but made permanent instead. Above it sits Claude Opus 5 ($5/$25) and the newer Claude Opus 5.5 ($4/$20), which is both more capable and cheaper than the model it replaced. At the budget end, Claude Haiku 4.5 runs $1/$5. A meaningful upgrade since 2025: models from Claude 4.6 onward include a full 1-million-token context window at standard pricing, instead of charging a premium for long context. Access has also expanded beyond Claude.ai and the API to Amazon Bedrock, Google Cloud Vertex AI, and now Microsoft Foundry. Subscriptions: Free, Pro ($20/month), and two Max tiers ($100 for 5x usage, $200 for 20x).
Google: Gemini 3.1 Pro leads a four-model family
Gemini 1.5's 2M-token context window was the headline feature in 2025; Google has since split the lineup into four tuned variants rather than one general model. Gemini 3.1 Pro (released February 19, 2026) is the current flagship, with a 1M-token context window and pricing in line with its Gemini 3 Pro predecessor ($2 input / $12 output per million tokens). Below it, Gemini 3.1 Flash-Lite targets cheap, fast tasks at $0.25/$1.50, and Gemini 3.5 Flash is tuned specifically for agentic coding. On the consumer side, Google AI Pro costs $19.99/month and bundles Gemini 3.1 Pro access with 1M-token context; Google AI Ultra was cut from $250 to $200/month for the 20x-usage tier, with a new $100/month entry-level Ultra tier offering roughly 5x Pro's limits.
DeepSeek: still the price disruptor, now fully open-weight
DeepSeek R1, the January 2025 model that originally shook up the market on cost, has been replaced by DeepSeek V4 (April 24, 2026) and a lighter DeepSeek V4-Flash, open-sourced under the MIT license on July 31, 2026. V4-Flash is a 284-billion-parameter mixture-of-experts model with only 13 billion parameters active per token, and both V4 variants support a 1-million-token context window with up to 384,000 output tokens. The pricing gap with closed-source competitors has, if anything, widened: DeepSeek-V4-Pro runs $0.66 per million input tokens and $1.98 per million output tokens at off-peak hours (01:00–04:00 and 06:00–10:00 UTC on weekdays), doubling during peak hours — a quirk none of the closed-source providers use. DeepSeek remains free to use through its own web app, with no subscription tier; the entire business model runs on API and self-hosting revenue.
xAI: Grok is no longer exclusive to X
The biggest structural change for Grok is access, not just capability. Grok-1's only access point in 2023 was X (formerly Twitter); the current flagship, Grok 4.7, is available through a standalone app and website at grok.com and via API, independent of any X account, alongside Grok 4.6 (released August 12, 2026). Both offer a 500,000-token context window and API pricing of $2.20 per million input tokens and $6.60 per million output tokens for prompts under 200,000 tokens, doubling above that threshold. Subscription options have also multiplied: X Premium ($8/month) and X Premium+ ($40/month, which includes SuperGrok access) sit alongside a separate SuperGrok ladder — Lite ($10), standard ($30), Plus ($100), and Heavy ($300, currently discounted to $99 for new subscribers). Worth noting for readers who just want the cheapest way in: X Premium+ at $40 is not automatically the better deal than standalone SuperGrok at $30, since they unlock different things.
Which one should you actually use
For general-purpose chat and everyday questions, Gemini 3.1 Pro or Claude Sonnet 5 give the best balance of price and quality. For coding specifically, Claude Opus 5.5 and DeepSeek V4-Pro are the two names developers bring up most — Opus for quality, DeepSeek for cost, especially if you're comfortable self-hosting the open-weight version. For tasks involving very large documents or codebases, Gemini 3.1 Pro, GPT-6 Astra, and Claude's 1M-token models are all viable; check actual pricing at that context length, since some providers charge more once you cross 200K tokens. For real-time or social-media-adjacent queries, Grok's X integration is still its clearest edge. And if budget is the deciding factor over everything else, DeepSeek's open-weight models are not a close second to anyone — they're the outlier.
What's still unsettled
A few things the original 2025 post didn't have to grapple with are now part of the decision. Regional data-residency pricing has arrived across providers — Anthropic and OpenAI both now charge roughly a 10% premium for guaranteeing US-only inference, which matters for regulated industries. Peak-hour pricing, pioneered by DeepSeek, adds real cost variability that a monthly-average price doesn't capture. And OpenAI's phased, access-gated rollout of GPT-6 Astra — reportedly tied to cybersecurity risk review — is a new pattern for releasing a frontier model that the other four companies haven't adopted, and it's not yet clear whether it becomes standard practice across the industry.
None of this lineup will look the same by early 2027 — it didn't hold from 2025 to now, either. We'll keep this comparison updated as the numbers move. What's your pick for daily use right now? Let us know in the comments.
-EditorZ
Post a Comment