OpenAI

Key Terms

• Token-based billing - charges based on the number of input, output, and cached tokens processed. One token is roughly 4 characters or 0.75 words. Codex credits are consumed based on per-model API token rates for Business and new Enterprise customers. Source: Openai – Pricing
• Prompt caching - stores frequently used prompt prefixes to cut latency and cost. Cached input tokens cost 90% less than regular input (e.g. $0.50/MTok cached vs $5.00/MTok standard for GPT-5.5). GPT-5.6 adds explicit cache breakpoints with a 30-minute minimum cache life, bills cache writes at 1.25x the uncached input rate, and keeps the 90% read discount. Source: OpenAI – Pricing , OpenAI – Previewing Gpt 5 6 Sol
• Batch API - asynchronous processing at 50% of standard rates, with results within 24 hours. GPT-5.5 Batch is $2.50/$15.00 per MTok. Source: OpenAI – Pricing
• Flex processing - best-effort processing at 50% of standard API rates, with possibly higher latency and lower availability. Source: OpenAI – Flex Processing
• Priority processing - guaranteed faster processing at 2.5x standard API rates. GPT-5.5 Priority is $12.50/$75.00 per MTok. Source: OpenAI – Pricing
• Context window - the maximum tokens a model can process in one conversation. GPT-5.5 supports 1M tokens via API and 400K in Codex for Pro/Enterprise. Source: OpenAI – Introducing Gpt 5 5
• Fast mode - generates tokens faster for a higher credit cost in Codex, for supported models. Source: Openai – Speed
• Credits - the core billing unit for Codex usage beyond included limits. Business and new Enterprise customers are billed based on API token rates mapped to credits. Source: Openai – Pricing
• GPT-5.3-Codex - a specialized coding model priced at $1.75/$14.00 per MTok via API. The standalone GPT-5.3-Codex entry no longer appears in the Codex credit rate card; it has been superseded in that card by GPT-5.3-Codex-Spark. Source: OpenAI – Pricing , Openai – Pricing
• GPT-5.3-Codex-Spark - a fast Codex model in research preview for ChatGPT Pro users only, not available via API, with a separate usage limit that may adjust based on demand. Source: Openai – Pricing
• GPT-5.6 Sol / Terra / Luna - the new GPT-5.6 naming system introduced June 26. The number identifies the generation, while Sol, Terra, and Luna identify durable capability tiers. Sol is the flagship ($5/$30 per MTok), Terra is the balanced tier ($2.50/$15), and Luna is the fast, low-cost tier ($1/$6). Source: OpenAI – Previewing Gpt 5 6 Sol
• max reasoning effort / ultra mode - new GPT-5.6 reasoning controls. max gives Sol the most time to reason, and ultra goes beyond a single agent by spinning up subagents to accelerate complex work. Source: OpenAI – Previewing Gpt 5 6 Sol
• Cyber Critical threshold - the danger level in OpenAI's Preparedness Framework above which a cyber-capable model is considered too risky to deploy. OpenAI states GPT-5.6 Sol does not cross it. Source: OpenAI – Previewing Gpt 5 6 Sol
• Data residency - processing data in specific geographic regions (US, EU, UK, JP, CA, KR, SG, IN, AU, UAE) with a 10% price uplift for eligible models released on or after March 5, 2026. Enterprise only. Source: OpenAI – Pricing

Latest Changes

Changes since the 2026-05 report.

• Verified (May watch-item): Pro promo multipliers expired. The Codex pricing page now describes Pro as "Choose 5x or 20x higher rate limits than Plus" at "From $100/month," with no 10x or 25x promotional tiers. Pro $100 reverts to 5x Plus usage and Pro $200 reverts to 20x Plus usage, exactly as the May report predicted. Source: Openai – Pricing
• Verified (May watch-item): GPT-5.6 is real and launched June 26. OpenAI previewed the GPT-5.6 series as three models, Sol (flagship), Terra (balanced), and Luna (fast/cheap), confirming the late-May Reddit sighting of "GPT-5.6" in the Codex UI. Source: OpenAI – Previewing Gpt 5 6 Sol
• Major event (governance): GPT-5.6 release is staggered at the request of the US government. The Trump administration asked OpenAI to start with a limited preview for a small group of "trusted partners" whose participation has been shared with the government, before broader release. OpenAI says it does not believe this should become the long-term default. This is the same mechanism that forced Anthropic to suspend Fable 5 and Mythos 5 on June 12. Sources: OpenAI – Previewing Gpt 5 6 Sol , Washingtonpost – Openai Says Us Government Will Vet Users Its Latest Ai Model , Bloomberg – Trump Administration Asks Openai To Stagger Release Of Ai Model
• Verified (May watch-item): GPT-5.5 API pricing unchanged. GPT-5.5 remains $5.00 input / $0.50 cached input / $30.00 output per MTok (long context $10.00/$1.00/$45.00), identical to the May report. The platform pricing page still labels "Latest: GPT-5.5" and does not list GPT-5.6 in the standard pricing table. Source: OpenAI – Pricing
• New model (limited preview): GPT-5.6 Sol/Terra/Luna pricing announced. Per MTok: Sol $5 input / $30 output, Terra $2.50 / $15, Luna $1 / $6. GPT-5.6 also introduces a new max reasoning effort and an ultra mode that uses subagents. During the preview the models are available only to a select group through the API and Codex. Source: OpenAI – Previewing Gpt 5 6 Sol
• New caching behavior (GPT-5.6): More predictable prompt caching with explicit cache breakpoints, a 30-minute minimum cache life, cache writes billed at 1.25x the uncached input rate, and cache reads keeping the 90% discount. Source: OpenAI – Previewing Gpt 5 6 Sol
• Benchmark claim (GPT-5.6): OpenAI says Sol sets a new state of the art on Terminal-Bench 2.1 (command-line workflows), beats GPT-5.5 on GeneBench v1 (genomics/biology) while using fewer tokens, is competitive with Mythos Preview on ExploitBench using about one-third of the output tokens, and shows gains on ExploitGym as reasoning increases. Exact numeric scores were published as charts only, not extractable text. Source: OpenAI – Previewing Gpt 5 6 Sol
• Safety claim (GPT-5.6): OpenAI spent over 700,000 A100-equivalent GPU hours on automated red teaming for universal jailbreaks. GPT-5.6 Sol does not cross the Cyber Critical threshold and did not autonomously produce a functional full-chain exploit in Chromium/Firefox tests, though it found bugs and exploitation primitives. Source: OpenAI – Previewing Gpt 5 6 Sol , Deploymentsafety – Gpt 5 6 Preview
• Feature GA: Codex Remote reached general availability on June 25. Users can start or continue work on a connected Mac or Windows host from the ChatGPT mobile app, with authenticated one-to-one QR pairing. A new DigitalOcean plugin lets Codex provision a Droplet as a remote workspace. Source: Openai – Changelog
• Credit-rate-card change: The Codex credit rate card dropped the standalone GPT-5.3-Codex row (formerly 43.75 / 4.375 / 350 credits per MTok) and now lists only GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5.3-Codex-Spark (research preview, no published rate), and a split GPT-Image-2 with separate image (200/50/750) and text (125/31.25/250) rows. Source: Openai – Pricing
• Limit-table tweak: The Pro 5x GPT-5.5 local-message range is now 75-400 per 5h (was 80-400 in May). Pro 20x (300-1,600) and Plus (15-80) GPT-5.5 ranges are unchanged. Cloud tasks remain "Not available" for all consumer/Business tiers on the live page. Source: Openai – Pricing
• Promo (referrals): From June 11 through June 24, 2026, eligible Plus and Pro users could invite up to three friends; when an invitee sent their first Codex message, both received a banked rate-limit reset usable for 30 days. Source: Openai – Pricing
• CLI releases: Codex CLI shipped 0.142.0 (June 22, with /usage earned-reset redemption, configurable rollout token budgets, multi-agent delegation config, indexed web-search mode, scheduled UTC time reminders), 0.142.1/0.142.2 (June 25, Windows/macOS system proxy support, MCP tool-search by default, safety-buffering UI metadata), 0.142.3 (June 26, maintenance), and 0.142.4 (June 29, no user-facing changes). Source: Openai – Changelog
• Hardware/infra: OpenAI and Broadcom unveiled an LLM-optimized inference chip (codenamed Jalapeno) on June 24, relevant to long-run inference cost but with no direct Codex price impact announced. Source: OpenAI – Openai Broadcom Jalapeno Inference Chip
• Community/security signal: The long-open GitHub issue requesting a way to exclude sensitive files in Codex (openai/codex#2847) hit HackerNews on June 28 with 226 points and 143 comments, prompting an OpenAI engineer to point to Codex's existing (Beta) permission profiles. Source: News – Item , Openai – Permissions

Plans

Codex usage limits per 5h window (current, after promo expiry):

Notes: Local messages and cloud tasks share a 5h window. Additional weekly limits may apply. Enterprise/Edu users on flexible pricing have no fixed rate limits; usage scales with credits. Enterprise and Edu plans without flexible pricing have the same per-seat usage limits as Plus. Source: Openai – Pricing

GPT-5.6 in Codex: During the preview, GPT-5.6 models are available through Codex and the API only to "a select group of trusted partners and organizations." There is no published Codex usage-limit or credit-rate row for GPT-5.6 yet. Source: OpenAI – Previewing Gpt 5 6 Sol

API Pricing

Standard pricing (per 1M tokens), unchanged from May:

Specialized models (per 1M tokens):

GPT-5.6 preview pricing (per 1M tokens), announced June 26, not yet on the standard pricing page:

Source: OpenAI – Previewing Gpt 5 6 Sol

Batch pricing (50% discount) and Flex pricing (50% discount, same rates) per 1M tokens:

Priority pricing (2.5x standard) per 1M tokens:

Other pricing:

• Web search: $10.00 per 1,000 calls plus search-content tokens billed at model rates
• Containers (Hosted Shell / Code Interpreter): $0.03 (1GB) to $1.92 (64GB) per 20-minute session, 5-minute minimum
• File search storage: $0.10/GB per day (1GB free); file search tool call: $2.50 per 1,000 calls
• Data residency: 10% uplift for eligible models released on or after March 5, 2026
• Fine-tuning (o4-mini only): $100/hour training, $4/$16 per MTok inference ($2/$8 with data sharing); the fine-tuning platform remains closed to new users

Codex credit rates (Business and new Enterprise, credits per 1M tokens):

GPT-5.5 usage averages 5-45 credits per message. Source: Openai – Pricing

Source (overall): OpenAI – Pricing , Openai – Pricing

Model Performance / Benchmarks

GPT-5.5 benchmarks (from the April 23, 2026 launch, still the current GA numbers):

Source: OpenAI – Introducing Gpt 5 5

GPT-5.6 Sol benchmarks (June 26 preview): OpenAI claims Sol sets a new state of the art on Terminal-Bench 2.1 (succeeding Terminal-Bench 2.0 above), beats GPT-5.5 on GeneBench v1 while using fewer tokens, is competitive with Mythos Preview on ExploitBench using about one-third of the output tokens, and improves on ExploitGym as reasoning increases. No exact numeric scores were published in extractable text; the headline figures appear as chart images in the launch blog and in the GPT-5.6 Preview System Card PDF. OpenAI says it will share an expanded evaluation suite when the model is broadly available. Source: OpenAI – Previewing Gpt 5 6 Sol , Deploymentsafety – Gpt 5 6 Preview

Note: OpenAI previously deprecated SWE-bench Verified in favor of SWE-Bench Pro, arguing the former is saturated and Python-only, and has noted evidence of memorization on SWE-Bench Pro by other labs. Source: OpenAI – Introducing Gpt 5 5

Latest News

GPT-5.6 Sol/Terra/Luna Preview and Government-Vetted Release (June 26, 2026)

OpenAI began a limited preview of the GPT-5.6 series: Sol (flagship, $5/$30 per MTok), Terra (balanced, $2.50/$15), and Luna (fast/cheap, $1/$6). The launch introduces a new naming system where the number is the generation and Sol/Terra/Luna are durable capability tiers, plus a new max reasoning effort and an ultra mode that uses subagents. At the request of the US government, OpenAI started with a limited preview for a small group of trusted partners whose participation was shared with the government, before any broader release. The models are available during the preview through the API and Codex only to that select group. Sources: OpenAI – Previewing Gpt 5 6 Sol , Deploymentsafety – Gpt 5 6 Preview

US Government to Vet GPT-5.6 Access (June 25-26, 2026)

The Washington Post reported that the US government will decide who gets to use GPT-5.6, and Bloomberg, Reuters, and The Verge reported the Trump administration asked OpenAI to stagger the release. OpenAI says it does not believe this government access process should become the long-term default and is taking the short-term step because it believes it is the strongest path to broader availability in the coming weeks. This parallels the June 12 export-control directive that forced Anthropic to suspend Fable 5 and Mythos 5. Sources: Washingtonpost – Openai Says Us Government Will Vet Users Its Latest Ai Model , Bloomberg – Trump Administration Asks Openai To Stagger Release Of Ai Model , Theverge – Openai Will Delay Gpt 5 6 After Trump Administration Request

METR Predeployment Evaluation Published (June 26, 2026)

METR published a summary of its predeployment evaluation of GPT-5.6 Sol, an independent third-party read on the model ahead of release. Source: Metr – 2026 06 26 Gpt 5 6 Sol

Codex Remote Reaches General Availability (June 25, 2026)

Codex Remote GA lets users start or continue work on a connected Mac or Windows host from the ChatGPT mobile app, review progress, and approve actions, with authenticated one-to-one QR pairing. Connections used since June 8 stay paired; older inactive connections need to re-pair. A new DigitalOcean plugin provisions a Droplet as a remote workspace. Source: Openai – Changelog , Openai – Remote Connections

OpenAI and Broadcom Unveil Inference Chip (June 24, 2026)

OpenAI and Broadcom unveiled an LLM-optimized inference chip codenamed Jalapeno. This is an infrastructure/cost story with no announced direct Codex price impact. Source: OpenAI – Openai Broadcom Jalapeno Inference Chip

Daybreak Security Tools (June 22, 2026)

OpenAI launched Daybreak, described as tools for securing organizations, alongside continued investment in Codex Security (the security-scanning plugin and cloud product). Source: OpenAI – Daybreak Securing The World

GeneBench-Pro and Applied AI (June 23-30, 2026)

OpenAI introduced GeneBench-Pro (June 30), a genomics/quantitative-biology evaluation tied to GPT-5.6's biology claims, and published an applied-AI story about GPT-5 helping immunologist Derya Unutmaz solve a multi-year mystery (June 23). Sources: OpenAI – Introducing Genebench Pro , OpenAI – Gpt 5 Immunology Mystery

Codex Sensitive-Files Issue Surfaces on HackerNews (June 28, 2026)

The long-open GitHub issue requesting a way to exclude sensitive files from Codex (openai/codex#2847) reached 226 points and 143 comments on HackerNews. An OpenAI engineer (bolinfest) responded that Codex already has a Beta "permission profiles" feature for marking paths/globs readable, writable, or unreadable, and for fine-grained sandbox network proxy control. Sources: News – Item , Openai – Permissions

Community Signals

The GPT-5.6 government-vetting story dominated June

The single biggest OpenAI story on HackerNews in June was the government-vetted GPT-5.6 release. "U.S. government will decide who gets to use GPT-5.6" reached 1,182 points and 1,240 comments, and the companion "Previewing GPT-5.6 Sol" thread reached 1,133 points and 743 comments. The dominant themes were concern that frontier-model access is now subject to ad hoc US government approval, predictions that buyers outside the US will pivot to Chinese open models, and direct pushback on OpenAI for years of asking governments to regulate AI. Sources: News – Item , News – Item

Key quotes from the "U.S. government will decide who gets to use GPT-5.6" thread:

• quantumwoke: "This is for the preview period, but it's not a good sign. Opus 4.8 may be the last frontier model available to the masses..." News – Item
• NooneAtAll3: "\"government needs to step in and regulate ai\" \"wait, not like that\"" News – Item
• SkitterKherpi: "So them banning Fable for only non-Americans is what we non-Americans should expect to be the norm going forward? ... the rest of us have to side with China, not because we remotely like it, but because they don't [restrict]." News – Item
• pixelpoet: "Are these models still relevant for people outside the US? I get the impression we're stuck on GPT 5.5 and Opus 4.8 pretty much permanently now, and relying on Chinese models in future." News – Item
• razighter777: "I hope this doesn't become the new norm where government becomes the bottleneck for innovation in the AI space. ... There's been no public policy, executive [order] or legislative process" behind it. News – Item
• kmeisthax, quoting OpenAI's statement that this should not become the long-term default: "My brother in Christ, then why did you (and your competitors) spend years telling the [government to regulate AI]?" News – Item

Naming and pricing pushback on the Sol/Terra/Luna launch

In the preview thread, developers objected to the new naming scheme layered over the existing mini/nano/standard/pro tiers and questioned whether GPT-5.6 is a real generation jump.

• rvz: "Other than the worst naming I have ever seen (Sol / Terra / Luna), the pricing is still expensive ... Sol is $5 input / $30 output; Terra is $2.50 input / $15 output; Luna is $1 input / $6 output." News – Item
• loufe: "\"Next generation model\" If it was the next generation, why isn't it a major version change..?" News – Item
• meetpateltech: "Another model family, another naming scheme to get used to. Sol Ultra [approximately equals] Pro, Sol [approximately equals] Standard, Terra [approximately equals] Mini, Luna [approximately equals] Nano." News – Item
• mekpro: "We need more coding benchmark score. Not sure that winning terminalbench 2.1 alone is a clear win over Fable/Mythos yet." News – Item
• HyperL0gi, on pricing-tier churn: "GPT-5 mini costs $0.25/$2 and will be discontinued in December. GPT-5.4 mini costs $0.75/$4.5 and is supposed to be the replacement. GPT-5.4 nano costs $0.2/$1.25 and ... is not even close when you test it in real [use]." News – Item

Codex sensitive-files debate: the community says sandbox, OpenAI says permission profiles

The HackerNews thread for the long-open Codex sensitive-files issue (226 points, 143 comments) drew near-unanimous agreement that an .agentsignore-style blocklist cannot be a security boundary, and that real isolation requires a sandbox or VM. An OpenAI engineer then pointed to Codex's existing Beta permission profiles.

• pikseladam (issue author): "it has been a year and still it is not resolved" News – Item
• kstenerud: ".agentsignore is NOT a security tool. ... However, using it to prevent exposure of secrets would be a BIG mistake. ... For security, use a sandbox. Nothing else will do." News – Item
• nikhilsimha: "Files that codex and any other coding agent has access to, should be opt-in NOT opt-out. I think codex is not the right layer to solve this if you want a sane(one-click) UX." News – Item
• bolinfest (OpenAI, Codex): "We have had a solution for this built into Codex for several months now. It is marked \"Beta\" in the docs ... With permission profiles in Codex, you can: mark a path, glob, or meta variable (like :workspace_roots) readable, writable, or unreadable." News – Item

Failed sources

The Reddit JSON endpoints (old.reddit.com/r/OpenAI/...json) returned HTTP 403 for multiple June Codex and GPT-5.6 threads, as in prior months, so June Reddit sentiment could not be fetched programmatically and is not included. HackerNews coverage above was used instead. The Reddit LocalLLaMA thread "US Govt to individually approve who gets GPT 5.6" (145 points) was visible via the HN Algolia index but its comments could not be read via JSON.

Enterprise Readiness

Terms explained:

• SCIM - System for Cross-domain Identity Management. Lets accounts sync automatically between your identity provider (Okta, Azure AD) and ChatGPT Enterprise. Source: Openai – Admin Setup
• EKM - Enterprise Key Management. Customer-managed encryption keys so OpenAI cannot decrypt your data without your key. Source: Openai – Admin Setup
• IP indemnity - the provider covers your legal costs if its AI output infringes a third party's copyright. OpenAI's IP-indemnity posture is not documented on the Codex or ChatGPT pricing pages, unlike GitHub Copilot's explicit Copilot Copyright Commitment.

Transparency Gaps