AI Coding News

New: an enterprise section for team and contract plans

The new Enterprise section lists 13 self-serve team tiers next to 6 enterprise contracts: Claude, ChatGPT, Cursor, Devin, Augment and Warp. The Compare table shows which vendors list SSO, SCIM, audit logs, data residency, spend caps and VPC deployment, and filters by requirement. The Budget table gives seat prices, minimums and annual seat fees at 25, 100 and 500 seats, and downloads as CSV.

Impact: Each price carries a confidence label: published, reported or unpublished. Four of the six enterprise vendors publish no seat price and show Quote; ChatGPT Enterprise’s $45–$75 range and 150-seat minimum are reported, not published. A feature is marked only where the vendor lists it for that plan. Claude and ChatGPT rows link into the budget planner.

Source

Artificial Analysis moves its Intelligence Index to v5.0

Artificial Analysis has rebuilt its Intelligence Index as v5.0. It drops GDPval-AA, SciCode, AA-LCR and Humanity’s Last Exam, and adds FrontierCode 2.0, Terminal-Bench-Science and HLE-Diamond. CritPt is replaced by its v2 hard subset, and four categories (Knowledge Work, Coding, Science, General) now carry 25% each. The new scale is harder: Claude Opus 5.5 holds at 58, GPT-6 Astra rises from 53 to 57, and most other models fall 5 to 15 points. Kimi K3.1 Pro drops from 55 to 46 and DeepSeek V4.1 Flash from 45 to 32.

Impact: Our models table now shows v5.0 scores for 28 models, each at its highest measured reasoning effort. The 13 models Artificial Analysis has not yet re-scored keep their v4.3.2 score, marked with *, and should not be ranked directly against v5.0 numbers. Because v5.0 scores run lower, we cut the dataset’s inclusion floor from 25 to 10, so no model leaves the table. Cost per task for those 28 models is re-measured on the v5.0 run, which costs more to complete: Claude Opus 5.5 goes from $5.98 to $10.04 and DeepSeek V4.1 Flash from $0.26 to $0.60. The value-frontier chart plots only v5.0 models.

Source

How we'll measure the token market: the boken

From now on, our briefs on the AI market will quote token volumes and prices per boken, one billion tokens. Industry figures run to quadrillions of tokens, and at prices per million they turn into numbers too large to compare. Per boken they read more like barrels or kilowatt-hours. A boken price blends input and output tokens 3 to 1, the weighting Artificial Analysis uses to group models by price, so Claude Opus 5.5 costs $8,000 a boken and DeepSeek V4.1 Flash $262.50. The plans and models tables stay in prices per million, which is the scale of one developer's usage.

Impact: A boken measures volume, not work. Models count tokens differently and differ in quality, so every boken figure names its model. The unit's first outing is Part 3 of the US labs series.

Source

OpenCode v2.2.0

OpenCode v2.2.0 adds shareable Desktop plugins, experimental image and decision generation APIs for plugins and the SDK, and per-provider prompt cache TTLs. Three changes break existing setups: @opencode/ai's package root now exports the Promise API, the OpenCode GitHub Action is no longer published, and TUI plugins require OpenTUI 0.7.

Impact: 3 breaking changes: the @opencode/ai root API, removal of the GitHub Action, and OpenTUI 0.7 for TUI plugins.

Source

GPT-6.1 Terra in GitHub Copilot

GitHub added GPT-6.1 Terra to the Copilot model picker in VS Code, Visual Studio, Copilot CLI, the cloud agent, JetBrains IDEs, Xcode, Eclipse, GitHub Mobile and github.com. Copilot Business and Enterprise administrators control access through the model policy, and the rollout is gradual.

Impact: Available on Pro, Pro+, Max, Business and Enterprise at provider list pricing: $1.00 per million input tokens and $5.00 per million output.

Source

Command Code Desktop v0.1.58

Settings gained a Providers page where you add your own model providers and keys, and connect a ChatGPT or GitHub Copilot subscription. Kimi K3.1 Pro is now in the model picker.

Source

GPT-6.1 Terra ships at half the price of 5.6 Terra

OpenAI released GPT-6.1 Terra for computer use and agentic work, with standard API pricing of $1 per million input tokens and $5 per million output, 50% below GPT-5.6 Terra. Artificial Analysis measures it at 48 on the Intelligence Index, $0.40 per task and 90.9 output tokens per second, with 133.64 seconds to the first answer token at max reasoning. OpenAI says it matches GPT-6 Astra at one-tenth the token price and finishes tasks more than three times faster.

Impact: Agent work gets Astra-level output at a fraction of the token cost; available now to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and as gpt-6.1-terra in the API.

Source

K3.1 Pro Series is live

Kimi K3.1 Pro is the first model in the K3.1 series: a 3T-class model with a 1M-token context window, rolled out in Kimi Code, Kimi.ai, Kimi Work and the Kimi API. A Highspeed variant runs about 2–3× faster at the same quality, and the full weights are due by November 4.

Impact: 1M-token context; K3.1 Pro is included from Plus/Moderato, while 1M context and Highspeed need Pro/Allegretto.

Source

Pi 1.4.0

Pi 1.4.0 adds MiniMax-M3.1-Flash to the minimax and minimax-cn providers and moves the Xiaomi default to mimo-v2.6-pro after mimo-v2.5-pro left the catalog. The default GitHub Copilot model changed to gpt-6.1-sol after Copilot retired gpt-5.4.

Source

DeepSeek open-sources DeepInfer, its production inference engine

DeepSeek released the inference framework behind its production services under Apache 2.0. It serves DeepSeek-V4.1-Flash on NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs; the KV cache system, prefill/decode disaggregation and cross-replica load balancing are not included.

Impact: Read the brief: what shipped, what was held back, and the likely effect on third-party hosts.

Source

Command Code v1.85.0

Command Code v1.85.0 adds MiniMax M3.1 Flash to the CLI with a limited-time free variant, alongside the Beam 1.0 free model. The desktop build released the same day adds MiniMax M3.1 Flash, Nemotron 3.5 Super VL, Muse Spark Open and Beam 1.0 to the model picker.

Source

Agent SDK

Cursor's Agent SDK defines an agent as a folder of code, with instructions.md as the prompt, one typed tool per file, skills loaded on demand and MCP connections, then tests it with evals and deploys it to Cursor or your own infrastructure. Agents wake on pull request, comment and CI events from GitHub, GitLab and Bitbucket, on Slack mentions and DMs, on webhooks or on a cron schedule.

Source

GitHub Copilot models deprecated on October 19, 2026

GitHub retired six models across Copilot Chat, inline edits, ask and agent modes, and code completions, following the announcement on September 18. Suggested replacements are Gemini 3.8 Flash, GPT-6.1 Sol, GPT-6 Luna and Grok 4.7.

Impact: 6 models retired: Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini and Grok 4.5.

Source

MiniMax M3.1 Flash ships with tunable thinking depth

A new-generation 1M-context multimodal coding model, billed as a drop-in replacement for M3 at a lower price.

Impact: AA Intelligence Index 43 at $0.15/M input and $0.60/M output.

Source

DeepSeek V4.1 Flash upgraded to the 1017 release

DeepSeek V4.1 Flash was upgraded to DeepSeek-V4.1-Flash-1017 across the web, app and API. API calling method and pricing are unchanged, so `deepseek-flash` reaches the new model with no code change.

Impact: AA Intelligence Index 39 → 45; cost per task $0.49 → $0.26.

Source

DeepSeek opens its in-house search and fetch API

The search engine behind DeepSeek's web and app search is now available as a standalone API, billed per request rather than per token.

Impact: Search and Fetch priced at $0.75 per 1,000 successful requests each.

Source

Cloud Agent Canvases and Environment Improvements

Cloud agents can now build interactive canvases and share them by link, and the Environments API entered beta. Cloud agent environments now support up to 300 repositories.

Impact: Up to 300 repositories per cloud agent environment; large environments load 5x faster.

Source

Upcoming deprecation of selected GitHub Copilot models in mid-November

GitHub set November 16, 2026 as the retirement date for Claude Opus 5, GPT-5.6 Terra and Claude Opus 4.8, naming Claude Opus 5.5, GPT-6.1 Sol and GPT-6 Luna as replacements. Claude Opus 4.8 stays available to individual subscribers on annual plans.

Impact: 3 models retire on 2026-11-16.

Source

DeepSeek V4.1 Flash released with native multimodal support

The smallest model in DeepSeek's new architecture family, with native visual understanding. The previous-generation V4 Flash and V4 Flash Vision Exp were retired and now route to V4.1 Flash.

Impact: API prices cut on launch; legacy model names still accepted.

Source

DeepSeek moves to peak and off-peak API pricing

DeepSeek V4 Pro reached GA with native OpenAI Responses API support and low/high/max thinking-effort control, alongside the switch to peak and off-peak pricing at a 50% off-peak discount.

Impact: Peak hours are 01:00-04:00 and 06:00-10:00 UTC on weekdays; off-peak rates are half.

Source