CHOOSELY.AI
Monthly AI-tool changes report
Version 1.0
8 Verified AI-Tool Changes That Mattered in July 2026
Evidence-based analysis of the launches, access changes, pricing movement, and shutdown that materially affected AI-tool users during July.
- Reporting period
- July 1 to July 31, 2026
- Verification cutoff
- Verified through August 1, 2026
- Publication date
- Published August 1, 2026
- Author
- Choosely Editorial
- Evidence label
- Evidence-based analysis
The HTML report is the canonical record. Material corrections are dated and reflected across the HTML, PDF, and CSV.
Key findings
8
material changes verified
319
tools in the monitored catalog
87.5%
of verified changes were model-related
80%
largest verified price cut
Eight material changes cleared Choosely’s verification threshold from a monitored catalog of 319 AI tools.
July accelerated the shift from model scarcity to model competition. Major providers expanded access, launched more specialized tiers, moved deeper into coding and agent workflows, and sharpened price-performance competition. Choosely verified eight changes significant enough to affect user, buyer, developer, or administrator decisions.
The strongest movement came from OpenAI's GPT-5.6 launch and rapid repricing, Google's expanded Flash family, Kimi K3's service and open-weight release, Grok 4.5's expansion into API and work surfaces, and Anthropic's Opus 5 launch. Fable 5 returned after an access suspension, while GitHub Models demonstrated the opposite risk by completing a full retirement path.
The practical conclusion is not to rebuild a stack around every announcement. It is to update routing assumptions, compare completed-work economics, test the relevant alternatives, and remove access dependencies that no longer have a stable future.
July at a glance
| Primary change class | Changes | Share |
|---|---|---|
| Model or product launch | 5 | 62.5% |
| Availability or access change | 1 | 12.5% |
| Price change | 1 | 12.5% |
| Shutdown | 1 | 12.5% |
One primary class is assigned to each change for headline counting. Individual changes may also carry secondary pricing, capability, billing, or distribution consequences.
Coverage statement
This report covers material AI-tool and model changes that were effective or publicly released from July 1 through July 31, 2026 and verified by August 1, 2026. At the cutoff, Choosely's monitored catalog contained 319 AI tools.
Most monitored tools recorded no public change that met the report’s materiality and evidence standard during July.
A change qualified only when Choosely could establish a practical consequence, reliable first-party evidence, a relevant July date, the affected users or workflows, and a defensible new state. A prior state is published only when reliable evidence supports it.
This is not a census of every AI product worldwide. It does not cover every vendor, region, private beta, enterprise contract, or unpublished change. The figures describe Choosely's verified July dataset and should be cited with the reporting period, verification cutoff, and monitored-catalog scope intact.
The eight verified changes
Claude Fable 5
Claude Fable 5 returned to global availability
Anthropic announced the restoration on June 30, with Fable 5 access effective July 1. Mythos 5 also returned for a restricted set of approved US organizations. The operational detail matters: the restored model carried revised cyber safeguards, fallback behavior, and temporary usage terms rather than simply returning unchanged.
- Previous state
- Access to Fable 5 and Mythos 5 had been suspended for all customers on June 12 after a US export-control directive took immediate effect.
- New state
- Fable 5 returned globally on July 1 across Claude Platform, Claude.ai, Claude Code, and Claude Cowork. Eligible plans received up to 50% of weekly usage limits through July 7 before access moved to usage credits. Blocked Fable 5 requests can route to Opus 4.8, and Anthropic warned of additional false positives on routine coding and debugging requests.
- Who is affected
- Claude users, Claude Code and Cowork users, API developers, and teams that had removed Fable 5 from production routing during the suspension.
- Decision effect
- Reassess Fable 5 for high-complexity work, but test benign coding and security-adjacent requests for false positives and account for usage-credit consumption after the temporary allowance ended.
Announced or released . Effective . Verified .
Pricing basis: USD · Usage credits after temporary included-usage period
Grok 4.5
Grok 4.5 entered API, coding, and office workflows
The official source trail contains three relevant dates, so the report records them rather than forcing one disputed launch date. The material change was broader than another chat release: Grok 4.5 was positioned for coding, agentic tasks, engineering, and office-document workflows, while long-context pricing introduced a meaningful cost threshold.
- Previous state
- Grok 4.5 was not available through the SpaceXAI API or its later distribution surfaces.
- New state
- SpaceXAI release notes record API availability on July 8 at $2 input and $6 output per million tokens. A broader launch announcement followed on July 16 across Grok Build, Cursor, and the API, with EU API-console availability on July 17. Requests using at least 200,000 context tokens are priced at $4 input and $12 output per million tokens.
- Who is affected
- API developers, Cursor users, Grok Build users, engineering teams, and people applying AI to spreadsheets, documents, and long-context workflows.
- Decision effect
- Evaluate Grok 4.5 on the actual coding or document workflow, and include the higher long-context rates when requests may exceed 200,000 tokens.
Announced or released . Effective . Verified .
Pricing basis: USD · $2 input / $6 output per 1M tokens below 200K context; $4 / $12 at 200K or more
GPT-5.6 family
OpenAI moved GPT-5.6 to general availability
The launch created a clearer routing ladder between frontier work, everyday production tasks, and lower-cost high-volume execution. It also introduced a new billing mechanic for prompt-cache writes, which means launch economics cannot be summarized by input and output token rates alone.
- Previous state
- GPT-5.6 Sol, Terra, and Luna were available only through a limited preview for selected partners and organizations.
- New state
- The family became generally available across ChatGPT, Codex, and the OpenAI API with a staged global rollout. Launch rates were $5/$30 for Sol, $2.50/$15 for Terra, and $1/$6 for Luna per million input/output tokens. GPT-5.6 also introduced cache-write billing at 1.25 times uncached input rates while retaining a 90% discount for cache reads.
- Who is affected
- ChatGPT paid users, ChatGPT Work users, Codex users, API developers, and teams routing tasks across model tiers.
- Decision effect
- Re-evaluate OpenAI model routing by task complexity, latency, completed-work quality, cache behavior, and total cost. Use the later July 30 Terra and Luna rates for current comparisons.
Announced or released . Effective . Verified .
Pricing basis: USD · Per 1M input/output tokens; cache writes at 1.25x uncached input rate
Kimi K3
Kimi K3 made open-weight, long-context evaluation harder to ignore
The correct description is open-weight, not open-source in the OSI sense. The weights and code are governed by the Kimi K3 License, which includes commercial conditions for certain model-as-a-service businesses and very large products. The timing also matters: service access began before the full weights were published.
- Previous state
- Kimi K3 was unavailable across Kimi products and the API.
- New state
- Kimi K3 launched on July 16 across Kimi products and the API with native multimodality, a one-million-token context window, and pricing of $0.30 cached input, $3 uncached input, and $15 output per million tokens. Full weights followed on July 27 under the bespoke Kimi K3 License.
- Who is affected
- Developers, coding-agent users, open-weight adopters, and teams handling large codebases, long documents, or repeated cached context.
- Decision effect
- Add Kimi K3 to evaluations where deployment control, open weights, cache reuse, or long context have practical value. Review the license before commercial deployment and keep output cost and task reliability in the comparison.
Announced or released . Effective . Verified .
Pricing basis: USD · $0.30 cached input / $3 uncached input / $15 output per 1M tokens
Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber
Google expanded the Flash family for production agents
This was a three-model announcement, although only 3.6 Flash and 3.5 Flash-Lite were broadly available at launch. The key operator signal was routing: a higher-value workhorse, a throughput-focused tier, and a restricted cyber model, with lower output cost for 3.6 Flash but higher generation-to-generation pricing for Flash-Lite than 3.1 Flash-Lite.
- Previous state
- Gemini 3.5 Flash and Gemini 3.1 Flash-Lite were the prior production baselines for these roles.
- New state
- Google introduced Gemini 3.6 Flash at $1.50 input and $7.50 output per million tokens, Gemini 3.5 Flash-Lite at $0.30/$2.50, and the restricted Gemini 3.5 Flash Cyber model for a trusted-partner CodeMender pilot. Google reported that 3.6 Flash used 17% fewer output tokens than 3.5 Flash on the Artificial Analysis Index while lowering output price from $9 to $7.50.
- Who is affected
- Developers running production agents, coding workflows, multimodal document processing, high-volume classification, and computer-use tasks.
- Decision effect
- Re-test Google routing using completion quality, latency, retries, tool calls, and total task cost. Treat the 3.5 Flash-Lite launch price as a new-generation rate, not a repricing of the continuing 3.1 SKU.
Announced or released . Effective . Verified .
Pricing basis: USD · 3.6 Flash: $1.50/$7.50; 3.5 Flash-Lite: $0.30/$2.50 per 1M input/output tokens
Claude Opus 5
Anthropic launched Claude Opus 5 as a lower-cost frontier workhorse
The launch is the primary event. Copilot distribution is recorded as a consequence rather than counted separately, which keeps the report consistent with other model integrations. GitHub bills Opus 5 at provider API list price under usage-based billing and warns that enhanced cyber safeguards may block some security-adjacent requests.
- Previous state
- Claude Opus 4.8 was Anthropic's current Opus model for complex agentic coding and enterprise work.
- New state
- Claude Opus 5 launched at $5 input and $25 output per million tokens, matching Opus 4.8 and sitting at half Fable 5 list pricing. It supports a one-million-token context window, up to 128,000 output tokens, and a May 2026 reliable knowledge cutoff. It became available through the Claude API and major cloud platforms, and GitHub began a gradual Copilot rollout on the same day.
- Who is affected
- Claude API developers, enterprise teams, complex coding users, GitHub Copilot users, and administrators controlling model access and usage-based billing.
- Decision effect
- Test Opus 5 against Opus 4.8 and Fable 5 on complex, long-running work. In Copilot, review administrator enablement, provider-list billing, and cyber safeguard behavior before broad rollout.
Announced or released . Effective . Verified .
Pricing basis: USD · $5 input / $25 output per 1M tokens; Copilot usage billed at provider API list price
GitHub Models
GitHub Models completed its retirement path
The July event was not the original decision to retire the service. It was the publication of the final timeline, two user-impacting brownouts, and the effective shutdown. That distinction matters because it shows both the earlier warning and the July migration deadline.
- Previous state
- GitHub closed Models to new customers on June 16 while existing customers retained the playground, API, model catalog, and BYOK access.
- New state
- On July 1, GitHub set July 30 as the full retirement date, scheduled brownouts on July 16 and July 23, and stated that the playground, model catalog, inference API, BYOK endpoints, and related UI would become unavailable to all customers.
- Who is affected
- Developers using GitHub Models for experimentation, model discovery, inference, or bring-your-own-key workflows.
- Decision effect
- Remove GitHub Models dependencies, confirm replacement endpoints and authentication, and compare Microsoft Foundry, GitHub Copilot, or another provider path for the specific workflow.
Announced or released . Effective . Verified .
GPT-5.6 Terra and GPT-5.6 Luna
OpenAI cut Terra and Luna prices and replaced Priority Processing
This is the report's only continuing-SKU repricing event. OpenAI also reduced how Terra and Luna usage counts against paid Codex and ChatGPT Work subscriptions, while subscription prices and quota budgets remained unchanged.
- Previous state
- Terra cost $2.50 input and $15 output per million tokens. Luna cost $1 input and $6 output. Priority Processing was the premium API service tier.
- New state
- Terra fell to $2 input and $12 output, a 20% reduction. Luna fell to $0.20 input and $1.20 output, an 80% reduction. Fast mode replaced Priority Processing for GPT-5.6 Sol, offering up to 2.5 times Standard speed at twice the price while remaining backward compatible with priority-tagged requests.
- Who is affected
- API developers, Codex users, ChatGPT Work users, and teams routing high-volume or latency-sensitive workloads.
- Decision effect
- Update every GPT-5.6 cost model based on launch rates. Retest Luna for background agent loops, Terra for everyday production work, and Fast mode only where lower latency justifies the 2x Sol rate.
Announced or released . Effective . Verified .
Pricing basis: USD · Terra: $2/$12; Luna: $0.20/$1.20 per 1M input/output tokens; Sol Fast mode at 2x Standard
Pricing and billing changes
A price event is a change to the published rate of a continuing model, product, or plan. Launch pricing and generation-to-generation differences are recorded inside launch events but are not counted as repricing events.
Continuing-SKU repricing
| Model | Old input | New input | Old output | New output | Change |
|---|---|---|---|---|---|
| GPT-5.6 Terra | US$2.50 | US$2.00 | US$15.00 | US$12.00 | -20% |
| GPT-5.6 Luna | US$1.00 | US$0.20 | US$6.00 | US$1.20 | -80% |
Generation-to-generation and billing context
- Gemini 3.6 Flash kept input at $1.50 and reduced output from Gemini 3.5 Flash's $9 to $7.50, while Google reported 17% fewer output tokens on the Artificial Analysis Index.
- Gemini 3.5 Flash-Lite launched at $0.30/$2.50, above 3.1 Flash-Lite's $0.25/$1.50.
- Claude Opus 5 launched at $5/$25, matching Opus 4.8 and sitting at half Fable 5's $10/$50 list price.
- GPT-5.6 introduced cache-write billing at 1.25x uncached input rates, and Fast mode replaced Priority Processing for Sol at 2x Standard pricing.
Category analysis
Seven of eight verified changes were model launches, model access, or model pricing events. Third-party distribution is normally folded into the parent model launch unless it creates a separate material migration, billing, or access consequence.
Stack implications
- 1
Re-run model routing
Separate high-complexity reasoning, everyday production, background agent loops, coding, and long-context work instead of applying one frontier model to every task.
- 2
Update cost assumptions
Any comparison using GPT-5.6 launch pricing is stale for Terra and Luna. Internal cost models and purchasing decisions should use the July 30 rates.
- 3
Compare completed-work economics
A lower token rate can still cost more when extra tool calls, retries, longer outputs, or manual correction are required.
- 4
Use distribution as a consolidation test
Claude Opus 5 inside GitHub Copilot may reduce the need for a separate coding surface for some teams, although usage-based cost and administrator controls still matter.
- 5
Audit access dependencies
Record which workflows depend on a provider, endpoint, playground, model alias, or third-party access layer, then identify a replacement before a retirement becomes urgent.
- 6
Review licenses before adopting open-weight models
Kimi K3 may be attractive for deployment control and long context, but its commercial terms need to be part of the decision.
The Change Brief
Get the week’s AI changes in one clear read
Pricing moves, tool launches, free-tier changes and practical stack updates, filtered for people who actually use these tools.
Stay ahead of AI without following it all day. We’ll send you what matters each week.
What to watch in August
Watchlist, not July event totals
- Model aliases and endpoint migrations scheduled to take effect in August
- Enterprise default-model policies that change access without a user decision
- Whether Luna's price cut changes high-volume agent routing in practice
- Whether Gemini 3.6 Flash reduces completed-task cost, not only token cost
- Whether Kimi K3 translates open-weight interest into reliable production use
- Migration outcomes following the GitHub Models retirement
Methodology
Event definitions
Choosely uses one stable primary class per event: model or product launch, availability or access change, price change, plan change, capability change, deprecation, shutdown, or policy change. An event can have secondary effects but is counted once.
Source standard
Material claims prioritize first-party vendor announcements, pricing pages, release notes, changelogs, support pages, and product documentation. Third-party reporting can identify a story, but it does not independently qualify a public event when first-party evidence should exist.
Distribution standard
Third-party model availability is normally folded into the parent launch unless it creates a separate material migration, billing, or access consequence.
Verification standard
The evidence must support the material claim, timing, affected users, and new state. A previous state is published only when reliable evidence supports it. Announcements made in July but taking effect in August belong in the watchlist unless the July announcement itself created the material user consequence.
Limitations
- The monitored catalog is a defined Choosely population, not the entire global AI-tool market.
- Regional, negotiated enterprise, private beta, and unpublished changes may not be represented.
- Vendor claims are attributed to vendor sources and are not presented as controlled Choosely testing.
Downloadable data
The downloadable CSV contains the eight verified event-level records and the fields required to interpret and reconcile them. It excludes private operational information and unpublished material.
© 2026 Choosely. All rights reserved. The dataset may be quoted and referenced with attribution. Redistribution, republication or commercial reuse of the complete dataset requires written permission from Choosely.
Corrections and version history
Corrections
Material corrections will be dated and explained. Stable event IDs remain attached to records that are corrected or superseded. Factual correction requests can be sent to info@choosely.ai.
Version history
- Version 1.0 - August 1, 2026
- Initial publication containing eight verified material changes. Material corrections will be dated and reflected across the HTML, PDF, and CSV.
Cite this report
Recommended citation
Choosely Editorial. 8 Verified AI-Tool Changes That Mattered in July 2026. Choosely, August 1, 2026. https://choosely.ai/reports/ai-tool-changes/2026-07
Preserve the monitored-catalog scope and verification cutoff when quoting an aggregate statistic. Vendor-specific facts should continue to cite the relevant first-party source. Choosely is the source for the cross-market aggregation, classification, and interpretation.
Final July verdict
July 2026 was a model-market month. The useful response is to review routing, update costs, test the most relevant alternatives, and remove dependencies that no longer have a stable future. Rebuilding the entire stack around every announcement would be a considerably less useful use of August.
Want the weekly version? Subscribe to The Change Brief →