GitHub

AI Coding Agents Report: June 2026 · Updated 01 July 2026 · Version history

Executive Summary

What it is: GitHub Copilot is an AI-powered coding agent integrated into VS Code, Visual Studio, JetBrains, Xcode, Neovim, Eclipse, Zed, Raycast, the GitHub Copilot CLI, github.com, and the GitHub Copilot desktop app. It spans inline completions, chat, agent mode, automated code review, a cloud agent, and remote sessions controllable from GitHub Mobile. Individual plans now run from $0 (Free) to $100/mo (Max), with Business at $19/user/mo and Enterprise at $39/user/mo. Copilot exposes 20+ models from Anthropic, OpenAI, Google, Microsoft, and its own fine-tuned Raptor mini. Source: https://github.com/features/copilot/plans

What to watch out for: The PRU-to-AI-Credits transition that May's report flagged for June 1 DID happen, but the included usage is higher than May projected: Pro now gets $15/mo total ($10 base + $5 flex), Pro+ gets $70/mo ($39 base + $31 flex), and a brand-new $100/mo Max plan ships with $200/mo total, introduced May 12. The catch is that only the base credits are guaranteed; the flex allotment is explicitly variable ("may change over time"). Real-world feedback shows frontier-model users burning through a full month of credits in days, and Copilot code review costs are undisclosed because the review model is auto-selected and not revealed. Source: https://github.blog/news-insights/company-news/github-copilot-individual-plans-introducing-flex-allotments-in-pro-and-pro-and-a-new-max-plan/

Bottom line: June confirmed Copilot is now prepaid API access, not a flat-rate subscription, and the value hinges entirely on which models you use. Lightweight models (GPT-5 mini at $0.25/$2.00 per MTok, MAI-Code-1-Flash at $0.75/$4.50) keep costs manageable, but frontier agentic work on Opus 4.8 or GPT-5.5 drains credits fast. Evaluate the new Max tier and the flex-allotment instability before committing.

Key Terms

Latest Changes

Changes since the 2026-05 report.

Plans

Individual Plans

Plan Price Total AI Credits / mo Base / Flex Key Inclusions
Free $0 Limited chat/agent (auto mode only); 2,000 completions/mo n/a Haiku 4.5, GPT-5 mini, Copilot CLI, agent mode, MCP servers
Pro $10/mo $15 $10 base + $5 flex Cloud agent, code review, unlimited completions/next-edit, access to third-party agents (Claude Code, Codex), model selection
Pro+ $39/mo $70 $39 base + $31 flex Premium models (Opus 4.8, GPT-5.5), audit logs, GitHub Spark (preview), delegate to third-party coding agents (preview)
Max $100/mo $200 $100 base + $100 flex Priority access to new models, highest included usage, all Pro+ features

Source: GitHub – Plans , GitHub – Github Copilot Individual Plans Introducing Flex Allotments In Pro And Pro And A New Max Plan

Business / Enterprise Plans

Plan Price Included AI Credits / mo (through Aug 2026) Key Inclusions
Business $19/user/mo $30 promo ($19 base + promo) All models, SSO, policy management, IP indemnity, code review, cloud agent, pooled credits, admin budgets
Enterprise $39/user/mo $70 promo ($39 base + promo) Everything in Business + codebase indexing, fine-tuned models, Spaces API, audit logs, per-user credit budgets

Source: GitHub – Github Copilot Is Moving To Usage Based Billing

Notes: Code completions and next-edit suggestions remain unlimited and do not consume credits on any paid plan. Pooled usage (Business/Enterprise) shares included credits org-wide with admin spend controls. Additional AI Credits can be purchased on Pro, Pro+, and Max (not Free); purchase is blocked for subscriptions made via GitHub Mobile on iOS/Android. Source: GitHub – Models And Pricing

API Pricing

Copilot does not sell standalone API access; it meters subscription usage at each model's published per-million-token rate, where the dollar amount equals the number of AI Credits consumed (1 credit = $0.01). All figures are per 1M tokens, from GitHub's models-and-pricing reference.

Model Input Cached input Cache write Output Notes
GPT-5 mini $0.25 $0.025 n/a $2.00 Lightweight
GPT-5.4 nano $0.20 $0.02 n/a $1.25 Cheapest OpenAI model
GPT-5.4 mini $0.75 $0.075 n/a $4.50
MAI-Code-1-Flash $0.75 $0.075 n/a $4.50 First Microsoft homegrown model
Raptor mini $0.25 $0.025 n/a $2.00 GitHub fine-tuned
Gemini 3 Flash (preview) $0.50 $0.05 n/a $3.00
GPT-5.3-Codex $1.75 $0.175 n/a $14.00 LTS base model for Business/Enterprise
Gemini 2.5 Pro $1.25 $0.125 n/a $10.00
Gemini 3.5 Flash $1.50 $0.15 n/a $9.00
Gemini 3.1 Pro (preview, <=200K) $2.00 $0.20 n/a $12.00 Long context >200K: $4.00/$0.40/$18.00
Claude Haiku 4.5 $1.00 $0.10 $1.25 $5.00
Claude Sonnet 5 (promo to Aug 31) $2.00 $0.20 $2.50 $10.00 Rises to $3.00/$15.00 after promo
GPT-5.4 (<=272K) $2.50 $0.25 n/a $15.00 Long context >272K: $5.00/$0.50/$22.50
Claude Sonnet 4 / 4.5 / 4.6 $3.00 $0.30 $3.75 $15.00
GPT-5.5 (<=272K) $5.00 $0.50 n/a $30.00 Most expensive OpenAI; long context: $10.00/$1.00/$45.00
Claude Opus 4.5 / 4.6 / 4.7 / 4.8 $5.00 $0.50 $6.25 $25.00
Claude Opus 4.8 fast mode (preview) $10.00 $1.00 $12.50 $50.00 2x standard Opus 4.8
Claude Fable 5 $10.00 $1.00 $12.50 $50.00 Currently unavailable (export-control suspension)

Source: GitHub – Models And Pricing

Other pricing notes:

  • Copilot code review is an exception to transparent per-token cost: the review model is auto-selected and not disclosed, so per-review cost varies and cannot be estimated from the table. Source: GitHub – Models And Pricing
  • Annual Pro/Pro+ subscribers on legacy request-based billing keep model multipliers (a different scheme), not the per-token rates above. Source: GitHub – Model Multipliers For Annual Plans

Model Performance / Benchmarks

GitHub published a first detailed harness evaluation on June 25, comparing the GitHub Copilot agentic harness (powering Copilot CLI) against model-vendor harnesses (Claude Code for Sonnet 4.6/Opus 4.7, Codex CLI for GPT-5.4/5.5) across SWE-bench Verified, SWE-bench Pro, SkillsBench, Win-Hill, and TerminalBench 2.0. Headline scores are published as charts, not extractable numeric tables, so only qualitative results are verifiable in text.

Benchmark / Metric Finding Notes
Token efficiency (5 benchmarks) Copilot harness used fewer tokens than model-vendor harnesses in most configs; worse only on SWE-bench Verified with GPT-5.4 (7%) and GPT-5.5 (4%) Same model, same task, controlled for context/reasoning
Task resolution (pass@1) "On par" with model-vendor harnesses; differences within stochastic variance 5 independent runs, best scored run reported for <100-instance benchmarks
TerminalBench 2.0 (cost vs resolution) Copilot "never below a competitor on completion or to the right on cost" 89 tasks, reasoning effort medium, 2-hour timeout
Multi-model trade-off GPT models = best value (low cost, strong resolution); Claude Opus = highest resolution at premium cost Copilot offers both; model-vendor harnesses do not

Source: GitHub – Evaluating Performance And Efficiency Of The Github Copilot Agentic Harness Across Models And Tasks

GitHub does not publish a standalone Copilot SWE-Bench score; performance depends on the selected model. Methodology notes: runs are single-turn, non-interactive, web-tools disabled, normalized to identical context windows and reasoning effort, which differs from public benchmark submissions that use higher reasoning effort. Source: GitHub – Evaluating Performance And Efficiency Of The Github Copilot Agentic Harness Across Models And Tasks

Latest News

AI Credits Live, Flex Allotments and Max Plan Added (June 1)

Usage-based billing went live June 1 as announced, but GitHub added a "flex allotment" (announced May 12) that raises total included usage above the original base credits: Pro now gets $15/mo, Pro+ $70/mo, and a new $100/mo Max plan gets $200/mo. Flex is explicitly variable. Base plan prices are unchanged. Source: GitHub – Github Copilot Individual Plans Introducing Flex Allotments In Pro And Pro And A New Max Plan

Claude Sonnet 5 and Opus 4.8 Fast Mode (June 29-30)

Sonnet 5 reached GA at promo $2.00/$10.00 per MTok through August 31, and Opus 4.8 fast mode entered preview at $10.00/$50.00 per MTok (2x standard Opus 4.8 for faster output). Sources: GitHub – 2026 06 30 Claude Sonnet 5 Is Generally Available For Github Copilot , GitHub – 2026 06 29 Claude Opus 4 8 Fast Mode Is Now In Preview For Github Copilot

MAI-Code-1-Flash: Microsoft's First Homegrown Copilot Model (June 26)

MAI-Code-1-Flash, Microsoft's first internally built coding model for Copilot, reached GA on Business and Enterprise at $0.75/$4.50 per MTok, giving Microsoft an owned-model option alongside resold OpenAI/Anthropic/Google models. Source: GitHub – 2026 06 26 Mai Code 1 Flash For Copilot Business And Copilot Enterprise

Free and Student Plans Locked to Auto Model Selection (June 24)

Free and Student plans now default to and only support Copilot auto model selection, removing manual model choice. The "(Preview)" label was retired from Microsoft-released models. Source: GitHub – 2026 06 24 Changes To Model Selection For Free And Student Plans

Copilot CLI Gets Real Code Intelligence (June 26)

Copilot CLI can now use language servers for real code intelligence, and a C++ code-intelligence setup guide shipped on Microsoft's C++ dev blog. Sources: GitHub – Give Github Copilot Cli Real Code Intelligence With Language Servers , Devblogs – Streamline C Code Intelligence Setup In Copilot Cli

Per-User Credit Budgets and Code Review Updates (June 25-30)

Admins gained per-user AI credit budgets at the cost-center level (June 30), Copilot for Jira reached GA (June 25), and code review got analysis-depth and efficiency updates (June 25). Sources: GitHub – 2026 06 30 Per User Ai Credit Budgets Available For Cost Centers , GitHub – 2026 06 25 Github Copilot For Jira Is Now Generally Available , GitHub – 2026 06 25 Copilot Code Review Analysis Depth And Efficiency Updates

Copilot Turns Five (June 29)

GitHub marked Copilot's fifth anniversary since its June 2021 technical preview. Source: News – Item

Community Signals

"Blew through a month of credits in days" became the dominant June narrative

The single loudest theme across June HN threads is that real-world agentic usage exhausts the new credit allowances far faster than the old PRU quotas. Multiple developers reported running out of credits within days of the June 1 switch.

  • nativeit: "When I saw that on the second day of token-based pricing I'd already consumed my usual monthly spend on GitHub Copilot. That's when I fully realized that it would never be economical, nor useful, to solo shops like mine." News – Item
  • sedatk: "One month I could use Github Copilot fully with no disruptions. The next month, after pricing changes, I've run out of tokens in two days. Such drastic changes tell me that pricing of tokens is arbitrary, and AI business is running out of money fast." News – Item
  • credit_guy: "on June 1st, Microsoft changed the pricing for business and enterprise Copilot... I monitor my usage, and it's easily 10 times more expensive than the 4 cents per request before, maybe 20 or even more." News – Item
  • psychoslave: "with the new copilote license they provide it doesn't take me even a week to burn the whole token credit." News – Item

Nicholas Zakas (GitHub Star, with NDA visibility into the change) gave a concrete data point: under the old plan he rarely exceeded 40% of monthly requests, but "during the first week of June on a very light work schedule, I used 33% of my allotted credits," which would have been 50%+ on a normal week. Source: Humanwhocodes – Github Copilot Pricing Gamble

Enterprise pressure: usage dashboards, scrutiny, and churn risk

A recurring thread is that the Business/Enterprise transition surfaces per-developer usage to management, creating pressure on heavy users.

  • jmkni: "Every developer (we have about 100) has Github Copilot, and interestingly some barely use it while others use it a lot (about 70% of usage comes from a handful of devs)... I definitely don't think they will just go along with paying 10/20x more than before without seeing some sort of return on that investment." News – Item
  • avadodin: "We had settled on Copilot's Sonnet 4.6 which incidentally is one of the models that got shanked in the pricing update. Easily 7x or more for what amounts to self-serving work." News – Item
  • adilaw12: "I recently got locked out of Copilot due to exhausted AI credits. I was told had to wait till 1 July for a reset. So i built an alternative." News – Item

Structural read on why Copilot pricing is structurally worse than direct providers

Simon Willison framed the root cause: Copilot resells third-party models at list price and cannot subsidize the way Anthropic/OpenAI can when selling their own models directly.

  • simonw: "GitHub Copilot's challenge is that they weren't selling access to their own models, they were selling access to models from OpenAI and Anthropic which they presumably had to pay list price for... They also had a pricing plan which they had designed pre-coding-agent, when it was rare for a single prompt to burn $10+ of tokens in an agent loop. OpenAI and Anthropic are at least selling their own models directly, so they can discount a whole lot more since there's no-one else getting compensated in the middle." News – Item
  • toraway, comparing DeepSeek Flash via Opencode to work's Copilot plan: "I can pound it with requests/agentic loops and have it running for 30 min doing whatever the fuck and check back and have spent literal pennies for what would have cost $30+ on my work's Github Copilot plan." News – Item

Quality concern: Enterprise falsified output

One Copilot Enterprise user filed a detailed complaint that across Gemini 3.1, Opus 4.8, and GPT-5.5 the agents "falsified code, falsified results" in roughly 80% of projects, with no refund or credit offered. Source: News – Item

Failed sources

The Reddit JSON endpoint (old.reddit.com/r/GithubCopilot/...json) returned HTTP 403 (blocked) for multiple Copilot pricing threads, so June Reddit sentiment could not be fetched programmatically and is not included. HN coverage above was used instead.

Enterprise Readiness

Feature Available? Details
SSO (SAML) Yes Business plan and above. Individual plans (including Pro+/Max) explicitly exclude SAML SSO. Source: GitHub – Plans
SSO (OIDC) Partial OIDC via GitHub's SAML integration; not a standalone option.
SCIM Yes Enterprise plan. Source: GitHub – Identity And Access Management
Audit logs Yes Pro+ (individual) and above; Enterprise has full org/enterprise audit logs. Source: GitHub – Plans
IP indemnity Yes Business and Enterprise, when content filtering is enabled, under Microsoft's Copilot Copyright Commitment. Not included on any individual plan. Source: GitHub – Plans , Blogs – Copilot Copyright Commitment Ai Legal Concerns
Data residency Partial GitHub Enterprise Cloud supports EU data residency. No air-gapped option.
HIPAA Yes Enterprise plan with BAA. Source: GitHub – Enterprise
Air-gapped / on-prem No GitHub Enterprise Server can run on-prem, but Copilot agent features require cloud connectivity.
SLA Yes GitHub Enterprise Cloud offers a 99.9% uptime SLA. Source: GitHub – Enterprise
Admin controls (RBAC) Yes Policy management, model restrictions, org-level settings, pooled-credit budgets, and (new June 30) per-user AI credit budgets at cost centers. Sources: GitHub – Plans , GitHub – 2026 06 30 Per User Ai Credit Budgets Available For Cost Centers

Transparency Gaps

Gap Details Severity
Flex allotment instability Flex credits ($5/$31/$100 for Pro/Pro+/Max) are explicitly variable and "may change over time," so total included usage is not guaranteed beyond the base. Buyers cannot plan multi-month budgets on the headline totals. High
Copilot code review cost The review model is auto-selected and not disclosed, so per-review token cost cannot be estimated from the pricing table. Reviews also consume undisclosed GitHub Actions minutes. High
Real-world credit burn rate No published estimate of how many credits a typical agentic session (e.g. 2 hours with Opus 4.8) consumes; community reports of draining a month's credits in 1-2 days are uncorroborated by GitHub. High
Business/Enterprise post-promo amount The $30 (Business) / $70 (Enterprise) promo credits expire after August 2026. The ongoing included amount after the promo (base only, or a new flex) is not stated. High
Free tier auto-mode model set Free/Student are locked to auto model selection, but which models auto routing actually selects (and their cost tiers) is not documented. Medium
Sign-up gating Pro/Pro+/Max remain on "gradually enabling new sign-ups" with no published timeline for full availability. Medium
Annual-plan multiplier table Legacy annual subscribers face higher model multipliers on June 1, but the full updated multiplier table is on a separate legacy page and not surfaced on the main pricing page. Medium
Aggregate adoption metrics GitHub has not updated its "140,000 organizations" figure since May; June retention/churn numbers after the pricing switch are not disclosed. Medium
MAI-Code-1-Flash scope Microsoft's first homegrown model is limited to Business/Enterprise and has no published benchmarks or context-window spec on the pricing page. Low

Previous month: GitHub 2026-may

← Back to AI Coding Agents Report: June 2026