T-Minus Signal
What's New in AI
Claude API Version 2026-10-21 Drops temperature, top_p, and top_k
Anthropic released API version 2026-10-21 on October 21, 2026. Requests on that version that send temperature, top_p, or top_k return a 400 error.
Why it matters
Pinning the new version is a code change, not a header bump. Search your codebase for sampling parameters before you upgrade, or the first request fails.
MiniMax M3.1 Flash Arrives as a $0.15 / $0.60 Open-Weight Model
MiniMax released M3.1 Flash on October 19, 2026 and describes it as an ultra-efficient open-weight model. API pricing is $0.15 input and $0.60 output per million tokens, with cache reads at $0.015.
Why it matters
The price matches Mistral Small 4 and undercuts DeepSeek-V4.1-Flash at peak. Its parameter count and license were not stated consistently across MiniMax's pages on October 23, 2026, so read the model card before you self-host.
DeepSeek-V4.1-Flash-1017 Takes Over deepseek-flash, With MIT Weights
DeepSeek released DeepSeek-V4.1-Flash-1017 on October 17, 2026. It now serves the deepseek-flash API model, supports vision, and its weights are on Hugging Face under the MIT License. Prices did not change: $0.15 to $0.30 input and $0.60 to $1.20 output per million tokens, off-peak to peak.
Why it matters
A permissive license plus a 1M-token context window at this price makes it worth testing against your paid default. The old deepseek-v4-flash names still work but now point to the new build, so rerun your evals.
Meta Releases Muse Spark Open Weights
Meta published the Muse Spark Open weights on October 15, 2026: 816B total parameters, 42B active, and a 1M-token context window, under a Muse Model License with limited commercial terms.
Why it matters
Open weights are not the same as an open license. Read the commercial terms before choosing it over MIT or Apache 2.0 models such as DeepSeek-V4.1-Flash or Qwen3.8-27B.
GPT-5.5 Is Retired From ChatGPT, ChatGPT Work, and Codex
OpenAI retired GPT-5.5 from ChatGPT, ChatGPT Work, and Codex on October 14, 2026. In Codex with ChatGPT sign-in, the change took effect at noon PDT, with GPT-6.1 Sol as the replacement on paid plans and GPT-6 Luna on Free and Go. API access was not affected.
Why it matters
If a workflow pinned GPT-5.5 in the app, re-test it on the replacement model. In the API, gpt-5.5 still costs $5 / $30 per million tokens and had no shutdown date listed as of October 23, 2026.
Gemini 4 Argon Becomes Google's Flagship at $2 / $10, Then $4 / $20
Gemini 4 Argon became generally available on October 13, 2026. API pricing is introductory: $2 input and $10 output per million tokens through November 30, 2026, then $4 / $20 from December 1, 2026. It is not on the free API tier. GitHub added it to Copilot the same day.
Why it matters
Budget at the $4 / $20 rate, not the launch price. As listed on October 23, 2026, it ranked first on the LMArena text leaderboard (marked preliminary).
X Subscription Bundles Grok, Cursor Pro, and X Premium+
xAI launched X Subscription on October 12, 2026 with Plus ($8/month), Premium ($30), Super ($100), and Ultra ($200) tiers. Premium includes SuperGrok, Cursor Pro, and X Premium+ in one usage pool shared by Grok and Cursor. Cursor's own pricing page now sells these tiers through X Subscription Corp.
Why it matters
Cursor buyers now pick an X Subscription tier. Existing Cursor, SuperGrok, and X Premium plans continue unless you change or cancel them, and list prices exclude sales tax.
Meta Launches Muse Wonder on Its Model API at $4 / $20
Meta launched Muse Wonder on October 12, 2026 on the Meta Model API and Muse Code, at $4 input and $20 output per million tokens with a 1M-token context window. A larger Muse Wonder Full is available only through Muse Code subscriptions.
Why it matters
Muse Wonder lists the same price as Claude Opus 5.5. As listed on October 23, 2026, Muse Wonder (Max) was listed third on the LMArena text leaderboard.
Google Stops Selling New Gemini Code Assist Subscriptions
From October 9, 2026, new Gemini Code Assist Standard and Enterprise subscriptions can no longer be purchased. Existing subscriptions auto-renew through 2026 and end in 2027, and Google points customers to Antigravity.
Why it matters
Teams on Code Assist should plan the Antigravity move now. Antigravity Business starts at $30 per seat per month through Google Cloud, or pay-as-you-go with no seat fee.
Google Deprecates Gemini 3.7 Flash and Gemini 3.5 Flash
Google deprecated Gemini 3.7 Flash and Gemini 3.5 Flash on October 8, 2026. Requests to them are routed to Gemini 3.8 Flash and Gemini 3.6 Flash.
Why it matters
Auto-routing keeps calls working but changes the model behind them. Gemini 3.8 Flash costs $0.75 / $3.75 per million tokens through December 31, 2026, then $1.50 / $7.50.
Claude Haiku 5.5 Starts at $0.10 / $0.50 per Million Tokens
Anthropic released Claude Haiku 5.5 on October 7, 2026. It costs $0.10 input and $0.50 output per million tokens for prompts up to 100K tokens, and $0.50 / $2.50 above that, with a 1M-token context window. The same day, the Sonnet 5.5 cache-read price fell to $0.10.
Why it matters
Haiku 5.5 is now the cheapest current Claude model and costs less than Haiku 4.5 ($1 / $5) at every prompt length. Haiku 4.5 is a legacy model that remains available.
GPT-6 Becomes the Model in ChatGPT Chat
On October 7, 2026, OpenAI rolled out GPT-6 with Intelligent UI in ChatGPT Chat. Plus, Pro, Business, and Enterprise use GPT-6 Sol, and the Pro reasoning option uses GPT-6 Astra. The release notes say Free and Go use GPT-6 Luna from October 8.
Why it matters
OpenAI's GPT-6 help article still listed GPT-5.6 Luna for Free and Go on October 23, 2026, so check the model picker on your own account before relying on either page.
Mistral Large 4 Enters Public Preview With a 1M-Token Context
Mistral Large 4 has been in public preview since October 6, 2026: 1.05T total parameters, 52B active, a 1M-token context window, and $1.36 input / $4.18 output per million tokens.
Why it matters
As of October 23, 2026, Mistral's model card lists the open weights as coming soon. Until they ship, treat Large 4 as a hosted preview, not an open-weight option.
GPT-6.1 Sol Replaces GPT-6 Sol as OpenAI's Balanced Model
OpenAI released GPT-6.1 Sol on September 29, 2026 at $2 input and $10 output per million tokens, with cached input at $0.10, a 1.05M-token context window, and 128K max output. OpenAI now presents it between GPT-6 Astra and GPT-6 Luna.
Why it matters
Same list price as GPT-6 Sol, half the cached-input price. If you call gpt-6-sol in the API, test the move; ChatGPT Chat still runs GPT-6 Sol.
ChatGPT Pro Adds a $500 Tier
On September 29, 2026, OpenAI added Pro 500 at $500 per month alongside Pro 100 and Pro 200. Ultrafast speed is only on Pro 500, and all three Pro tiers bill monthly with no annual plan.
Why it matters
Pro is now a three-step ladder ($100, $200, $500), not a single $200 plan. Pick the tier by how often you hit weekly GPT-6 Pro limits.
Claude Sonnet 5.5 Launches at $2 / $10
Anthropic released Claude Sonnet 5.5 on September 28, 2026 at $2 input and $10 output per million tokens, with a 1M-token context window and 128K max output. Anthropic deprecated Claude Sonnet 4.5 on September 30 and named Sonnet 5.5 as its replacement.
Why it matters
Sonnet 5 is now a legacy model at the same price, so there is no cost reason to stay on it. As listed on October 23, 2026, Sonnet 5.5 placed second on the Artificial Analysis Intelligence Index.
Claude Sonnet 5 Keeps $2 / $10 as Its Standard Price
Anthropic's pricing page now lists Sonnet 5's $2 input and $10 output per million tokens, first announced as introductory pricing through August 31, 2026, as the standard price. The scheduled September 1 increase to $3 / $15 did not take effect.
Why it matters
Any budget built on the planned $3 / $15 rate overstates Sonnet 5 token cost by 50%. Sonnet 5 has since become a legacy model; Sonnet 5.5 lists the same $2 / $10 price.
Claude Opus 5.5 Becomes Anthropic's Recommended Starting Model
Anthropic released Claude Opus 5.5 on September 22, 2026 at $4 input and $20 output per million tokens, with a 1M-token context window and up to 128K output tokens. It is available on Pro, Max, Team, Enterprise, and the API.
Why it matters
Anthropic said Opus 5.5 costs 40% less to run than Opus 5 and is over 30% faster at default settings. Treat those as vendor claims and rerun your own evals before moving a production workflow. Claude Fable 5.1 ($10 / $50) remains the higher tier for demanding reasoning and long-horizon agent work.
GPT-6 Sol and GPT-6 Luna Split OpenAI's New Lineup by Cost
OpenAI released GPT-6 Sol ($2 input / $10 output per million tokens) for complex coding and agentic work and GPT-6 Luna ($0.10 / $0.50) for efficient, high-volume work on September 22, 2026. Both list a 1.05M-token context window.
Why it matters
OpenAI retired GPT-5.5 from ChatGPT, ChatGPT Work, and Codex on October 14, 2026. In the API, GPT-6.1 Sol (September 29) has since taken over the balanced slot from GPT-6 Sol at the same list price.
GPT-6 Astra Arrives as OpenAI's Most Capable Model
OpenAI released GPT-6 Astra on September 3, 2026 and describes it as its most capable model. API pricing is $10 input and $50 output per million tokens, with a 1.05M-token context window and up to 128K output tokens.
Why it matters
Astra is priced like Claude Fable 5.1 at $10 / $50. As of October 23, 2026, ChatGPT offers it in Chat as GPT-6 Pro on Pro, Business, and Enterprise, while Plus gets it only in Work and Codex.
OpenAI Cuts GPT-5.6 Sol API Pricing to $4 / $20
OpenAI reduced GPT-5.6 Sol to $4 input and $20 output per million tokens on August 21, 2026. The GPT-5.6 family (Sol, Terra, and Luna) launched on July 9, 2026.
Why it matters
A price cut on a model you have already tuned changes the math without a migration. Compare the new Sol rate against the September GPT-6 lineup before moving anything.
EU Digital Omnibus on AI Pushes Back High-Risk Deadlines
Regulation (EU) 2026/1744 entered into force on July 27, 2026. It moves standalone (Annex III) high-risk obligations from August 2, 2026 to December 2, 2027, and high-risk AI in regulated products (Annex I) to August 2, 2028.
Why it matters
The delay does not cover transparency. Article 50 obligations still apply from August 2, 2026, with a grace period until December 2, 2026 for machine-readable marking on generative systems already on the market. This is a summary, not legal advice.
Claude Opus 5 Listed With 1M Context and $5 / $25 Pricing
In July, Anthropic listed Claude Opus 5 as active with a 1M-token context window, up to 128K output tokens, and API pricing of $5 input and $25 output per million tokens. Anthropic now lists Opus 5 as a legacy model that remains available.
Why it matters
Opus 5.5 now lists a lower $4 / $20 price. Keep Opus 5 only where your own evals justify staying on it.
ChatGPT's July Plan Page Surfaced GPT-5.6 Access by Tier
In July 2026, the ChatGPT plan page listed limited GPT-5.5 Instant access on Free, GPT-5.6 advanced reasoning on Plus, and GPT-5.6 Sol Pro on Pro.
Why it matters
Plan pages change more often than model announcements. GPT-5.5 was retired from ChatGPT on October 14, 2026, so treat this entry as a July snapshot and check the current plan page before purchasing.
Gemini 3.6 Flash Led Google's Balanced Model Line in July
In July, Google documented Gemini 3.6 Flash as its most balanced model for speed and intelligence, with Gemini 3.5 Flash and Flash-Lite covering sustained agentic and lower-cost workloads. Google has since moved on: Gemini 3.8 Flash (generally available September 2) is its top Flash model, 3.6 Flash is labeled previous-generation, and 3.7 Flash and 3.5 Flash were deprecated on October 8, 2026.
Why it matters
For API buyers, the useful comparison is workload fit and standard token price, not a universal model ranking.
Grok 4.5 Is Listed at $2 Input and $6 Output per Million Tokens
xAI's Grok 4.5 release update lists standard API pricing of $2 per million input tokens and $6 per million output tokens.
Why it matters
xAI has since released Grok 4.7 (September 22, 2026), now its recommended model for code and chat at the same $2 / $6 headline price.
GPT-5.6 Introduces Sol, Terra, and Luna
OpenAI released the GPT-5.6 family on July 9, 2026: Sol as the flagship model, Terra as the balanced option, and Luna for lower-cost, latency-sensitive work.
Why it matters
The meaningful decision is no longer one GPT tier. Choose the model lane that matches your workflow and budget, then validate the actual product limits before committing.
GitHub Copilot Moves to Usage-Based Billing With AI Credits
GitHub moved Copilot to usage-based billing with AI Credits on June 1, 2026. Plan prices stayed the same: Pro $10, Pro+ $39, Business $19 per user, and Enterprise $39 per user.
Why it matters
The sticker price did not move, but the cost of a heavy agent month now depends on credit consumption. Track usage for a few weeks before changing seat tiers.
Mistral 3 Raises the Bar for Open Multimodal Models
Mistral introduced Mistral 3 as an open multimodal and multilingual model family, giving teams another serious open-model option for on-premise, private, and cost-sensitive deployments.
Why it matters
Open models are no longer just backup plans. They are becoming the governance and cost-control layer for teams that cannot send every workflow to closed APIs.
DeepSeek V4 Appears in Official Transparency Data
DeepSeek’s transparency center now lists DeepSeek V4 with an April 24, 2026 release date, keeping DeepSeek relevant in the low-cost and open-model conversation.
Why it matters
The key builder question is whether DeepSeek remains the cheapest capable API lane while frontier closed models push deeper into agents.
GPT-5.5 Turns ChatGPT Into a More Serious Work Agent
OpenAI released GPT-5.5 on April 23, 2026 and framed it around real computer work: coding, research, data analysis, documents, spreadsheets, software operation, and multi-tool task completion.
Why it matters
Update: OpenAI retired GPT-5.5 from ChatGPT, ChatGPT Work, and Codex on October 14, 2026. It remains available in the API at $5 / $30 per million tokens.
Workspace Agents Bring Shared ChatGPT Agents Into Team Workflows
OpenAI introduced workspace agents in ChatGPT for shared business workflows that can coordinate through tools, Slack-style processes, Codex-powered execution, and persistent team context.
Why it matters
This is the next version of custom GPTs: not personal toys, but shared agents that sit inside repeatable company workflows.
Gemini Deep Research Max Moves Research Agents Beyond Web Summaries
Google announced Deep Research and Deep Research Max for Gemini API workflows with Gemini 3.1 Pro, MCP support, private data connections, plan control, streaming, charts, and native visual outputs.
Why it matters
Research agents are becoming product infrastructure. The most interesting use case is not one report; it is background due diligence, market scans, and private-data research pipelines.
Canva AI 2.0 and Claude Design Expand the AI Creative Stack
Canva Create 2026 added Canva Offline, Pro Design updates, Affinity integrations, and Claude-connected creative workflows, while Anthropic launched Claude Design for prototypes, decks, and visual work.
Why it matters
Creators and founders now need a stack, not one design tool: Claude for structured visual work, Canva for production and distribution, and Affinity/Cavalry for deeper craft.
Gemini Workspace Updates Keep Moving AI Into Docs, Sheets, Slides, and Drive
Google’s March 2026 Workspace update expanded Gemini across Docs, Sheets, Slides, Drive, and related productivity surfaces for work grounded in the files teams already use.
Why it matters
Google’s advantage is distribution. If your work lives in Workspace, the AI layer increasingly arrives where the file already is.
GPT-5.4 Is Generally Available in GitHub Copilot
GitHub made GPT-5.4 generally available in Copilot in March 2026, strengthening the case for Copilot as the enterprise default for IDE-native AI assistance.
Why it matters
Copilot’s moat is not always model quality; it is distribution, policy, IDE support, and GitHub-native adoption.
Claude Sonnet 4.6 Makes 1M-Context and Computer Use More Practical
Anthropic introduced Sonnet 4.6 with stronger coding and computer-use behavior plus expanded long-context capabilities, including a 1M token context beta for qualified use cases.
Why it matters
For builders, Sonnet is the workhorse release: cheaper and faster than Opus, but increasingly capable enough for real agent loops.
Cursor Long-Running Agents Shift AI Coding From Chat to Delegation
Cursor expanded its long-running agents research preview for background coding tasks, making it easier to hand off implementation work instead of staying inside a single chat loop.
Why it matters
The new coding stack is editor plus agents plus review discipline. The winners will be tools that can keep working while developers do something else.
Claude Opus 4.6 Pushes Claude Further Into Premium Coding and Agent Work
Anthropic released Claude Opus 4.6 on February 5, 2026, positioning it as the high-end Claude model for difficult reasoning, software engineering, computer-use workflows, and agentic work.
Why it matters
Claude is no longer just the calm writing model. Its highest-value lane is now long-horizon technical work where context, judgment, and tool discipline matter.